{"state": {"task": "put beet close to but not into pot sink", "subtask": "put beet close to but not into pot sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4a95b21b72bf1d1b6c929ed7.png"}, {"type": "image", "uri": "media/bdv2fail/8c3dd778407922c566c46d97.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0000/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 0, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the red knob to the bottom of the burner.", "subtask": "move the red knob to the bottom of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ea8a3c30cd86e1a31796e6f6.png"}, {"type": "image", "uri": "media/bdv2fail/1909b954247d2d67fa8fa5e1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0001/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 1, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the yellow knife underneath of the yellow cloth", "subtask": "put the yellow knife underneath of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/16f82423b82e81746422b57d.png"}, {"type": "image", "uri": "media/bdv2fail/6275ca7f7387deadf13e9b0f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0003/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 3, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move an orange object to the top left of the table", "subtask": "move an orange object to the top left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70486cb351724ea807acf26b.png"}, {"type": "image", "uri": "media/bdv2fail/75e1f926e72ae04b4d3642c1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0005/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_11_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 5, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_11_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "place the brush near but not on the yellow cloth", "subtask": "place the brush near but not on the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/37c4dca1fb8c1328673f0976.png"}, {"type": "image", "uri": "media/bdv2fail/81860e40dd2ca4ba1b54c7a1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0006/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_10_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 6, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_10_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5a3f2d55fd39673d4e40df96.png"}, {"type": "image", "uri": "media/bdv2fail/9fbbca24a8a9fbb667d1f19c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0007/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 7, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 215, "role": "error_localization"}}
{"state": {"task": "move the pot to the right burner", "subtask": "move the pot to the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/570171b69a2550d9fc082a3b.png"}, {"type": "image", "uri": "media/bdv2fail/a8a765d5b203941192c226af.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0009/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 9, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the banana to the lower right of the table", "subtask": "move the banana to the lower right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86753477bfea123aa804406e.png"}, {"type": "image", "uri": "media/bdv2fail/86753477bfea123aa804406e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0016/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 16, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the green cloth to the edge of the table", "subtask": "move the green cloth to the edge of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6060f3e8ccc37d9355d1b79.png"}, {"type": "image", "uri": "media/bdv2fail/dbaab0384a8b1fdc1eb9c05e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0017/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train/18", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 17, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train", "episode_id": "18", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "turn faucet front to left.", "subtask": "turn faucet front to left.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9c3fd58e001fc1ff13d224aa.png"}, {"type": "image", "uri": "media/bdv2fail/a09d975c71db0eb7da9e95ee.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0019/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train/18", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 19, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "error_localization"}}
{"state": {"task": "put the red tomato on plate", "subtask": "put the red tomato on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/521abf6e09490588cc42d896.png"}, {"type": "image", "uri": "media/bdv2fail/df54d632021295075ba06308.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0024/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 24, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "put the silver pot on the orange towel", "subtask": "put the silver pot on the orange towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/90203c75a329c968da463e4a.png"}, {"type": "image", "uri": "media/bdv2fail/714039a9f66790bc1d9bce11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0026/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 26, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the blue brush to the upper left corner of the stove", "subtask": "move the blue brush to the upper left corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ee681ef9d8cb6bc96f60e4a5.png"}, {"type": "image", "uri": "media/bdv2fail/879ee3e55cc09c57871e37a8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0027/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 27, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the upper left of the stove", "subtask": "move the silver pot to the upper left of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/54a9552c030c81ceb14d760b.png"}, {"type": "image", "uri": "media/bdv2fail/856bd616d97ff1df4d7124d5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0028/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 28, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth to the right side of the burner", "subtask": "move the yellow cloth to the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4a39f48213acb40ccf4b537e.png"}, {"type": "image", "uri": "media/bdv2fail/4a39f48213acb40ccf4b537e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0030/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 30, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "3", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the blue plastic spoon to the left side of the blue fork.", "subtask": "move the blue plastic spoon to the left side of the blue fork.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/98d58ec9799f25aac67d0e37.png"}, {"type": "image", "uri": "media/bdv2fail/886314127642897041dfca29.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0031/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 31, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "3", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the blue spoon to the upper left of the red pepper", "subtask": "move the blue spoon to the upper left of the red pepper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/681f3d016aa9346d987aade2.png"}, {"type": "image", "uri": "media/bdv2fail/250740c61b1792ea2e94b3be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0032/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 32, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the green bell pepper in front of the drawer.", "subtask": "move the green bell pepper in front of the drawer.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9453cabac20bab1bc926b8dd.png"}, {"type": "image", "uri": "media/bdv2fail/95219641a8a28212a302ef30.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0034/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/34", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 34, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "34", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the red object to the bottom right of the stove", "subtask": "move the red object to the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ca03d1ed51d690c9da373fb.png"}, {"type": "image", "uri": "media/bdv2fail/9ca03d1ed51d690c9da373fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0037/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 37, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the black cord from right to left.", "subtask": "move the black cord from right to left.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4798d42389e06450c4a519dd.png"}, {"type": "image", "uri": "media/bdv2fail/e3947a77050cb45c5876be82.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0038/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 38, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "put the spoon in the green pepper", "subtask": "put the spoon in the green pepper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/416a77f5037762a3a284e9b7.png"}, {"type": "image", "uri": "media/bdv2fail/6818c3cb8b055b233438070d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0040/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 40, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the red fork on the upper left side of the table over the pot", "subtask": "move the red fork on the upper left side of the table over the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/18be481c9b51e172248e495d.png"}, {"type": "image", "uri": "media/bdv2fail/e17378ae2f5a1d7934a985e6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0043/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 43, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the silver pot in the right side of the mushroom", "subtask": "move the silver pot in the right side of the mushroom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/96aadcbe4b5f5e60bf978cdb.png"}, {"type": "image", "uri": "media/bdv2fail/426ceddab3b5b256bbe333c0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0046/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 46, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "place the yellow knife to the upper side of the blue dish towel.", "subtask": "place the yellow knife to the upper side of the blue dish towel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2d644706ba3f43a9678efd1e.png"}, {"type": "image", "uri": "media/bdv2fail/e275bfe7b6933a9d765a5d5e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0048/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_62_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 48, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_62_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "place the red spoon on the yellow cloth", "subtask": "place the red spoon on the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0bcbf6d75dbd334f7d462e82.png"}, {"type": "image", "uri": "media/bdv2fail/d1ff577c463d567c07e66568.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0049/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_109_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 49, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_109_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the silver frying pan to the right of the green towel.", "subtask": "move the silver frying pan to the right of the green towel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d48b0b79c11fbd1ce0ebbfb9.png"}, {"type": "image", "uri": "media/bdv2fail/a33bdde5f4ed17c5d274427c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0050/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 50, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the yellow knife to the left side of the left burner", "subtask": "move the yellow knife to the left side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c835af7b3f50870e2880974.png"}, {"type": "image", "uri": "media/bdv2fail/de985a24f417c71fc07959d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0051/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 51, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the blue object to the left edge of the table from the bottom right edge of the orange cloth.", "subtask": "move the blue object to the left edge of the table from the bottom right edge of the orange cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f0f4ca89a4eac09cb8d16a8.png"}, {"type": "image", "uri": "media/bdv2fail/acf4c124cb007ee22fb62960.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0052/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 52, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 249, "role": "error_localization"}}
{"state": {"task": "move the spatula to the right side of the silver pot from the upper part of the table", "subtask": "move the spatula to the right side of the silver pot from the upper part of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6ff11b85fb34c61b405aed0.png"}, {"type": "image", "uri": "media/bdv2fail/5311243f1d47f128a0f673d2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0053/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_02_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 53, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_02_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "put the red cup on the green pepper", "subtask": "put the red cup on the green pepper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a02ccaed04137e760423041f.png"}, {"type": "image", "uri": "media/bdv2fail/0962ac2f872e8c47534f515d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0055/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 55, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom right side of the stove", "subtask": "move the silver pot to the bottom right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/34cd516bb80990b2213577f0.png"}, {"type": "image", "uri": "media/bdv2fail/34cd516bb80990b2213577f0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0062/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/18", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 62, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "18", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "place the red object to bottom left edge of the cloth", "subtask": "place the red object to bottom left edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d97be847a74a4791e62d940.png"}, {"type": "image", "uri": "media/bdv2fail/a1fca23ea9e591b9384dd290.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0066/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 66, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc373be66bb13a2241110e5c.png"}, {"type": "image", "uri": "media/bdv2fail/cc373be66bb13a2241110e5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0067/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_06_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 67, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_06_train", "episode_id": "21", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 215, "role": "error_localization"}}
{"state": {"task": "move the silver pot next to the light green handle spatula.", "subtask": "move the silver pot next to the light green handle spatula.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c294208b6fde9d27d9506dc1.png"}, {"type": "image", "uri": "media/bdv2fail/08ff4d2f3451d36b3ba5cc03.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0071/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train/20", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 71, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train", "episode_id": "20", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the banana to the upper middle of the toy stove.", "subtask": "move the banana to the upper middle of the toy stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2a019097dd9a28e6ed69d6f9.png"}, {"type": "image", "uri": "media/bdv2fail/3226f8e26736e315615ac3c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0072/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/18", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 72, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom left corner of the burner", "subtask": "move the silver pot to the bottom left corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c90750c2ccb5c44fcd5f9de5.png"}, {"type": "image", "uri": "media/bdv2fail/d8710242fa889b3b3f52320a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0075/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 75, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put red pepper on plate", "subtask": "put red pepper on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6ad1a3c7322d7f9b7d118cd.png"}, {"type": "image", "uri": "media/bdv2fail/7a0a8de7f4dbb945917e87e9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0077/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 77, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "error_localization"}}
{"state": {"task": "move the yellow knife to the upper side of the blue cloth.", "subtask": "move the yellow knife to the upper side of the blue cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9ce29d894602fc9b44bea0e.png"}, {"type": "image", "uri": "media/bdv2fail/714d552e8f8084419258beb6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0078/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_108_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 78, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_108_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "moving the green spatula into the right side of the stove", "subtask": "moving the green spatula into the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf70ef1fb3814e12ee7d1a4f.png"}, {"type": "image", "uri": "media/bdv2fail/111b934ecc306d90bc23feec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0081/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 81, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put cup on plate", "subtask": "put cup on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7ef2a9e672d01087d4708ce4.png"}, {"type": "image", "uri": "media/bdv2fail/7ef2a9e672d01087d4708ce4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0084/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/27", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 84, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "27", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "turn faucet front to left.", "subtask": "turn faucet front to left.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/03b196810f1e2d3e9a3cb012.png"}, {"type": "image", "uri": "media/bdv2fail/65048c1595538fd5727b45b5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0085/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 85, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "error_localization"}}
{"state": {"task": "move the silver pot from the top right corner of the cook to the left burner.", "subtask": "move the silver pot from the top right corner of the cook to the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/051f24ca4e6f0d672d3560ca.png"}, {"type": "image", "uri": "media/bdv2fail/aca3a0b630d2b7fd14bfef3e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0088/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 88, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "move the blue fork to the upper right side of the stove next to the red object", "subtask": "move the blue fork to the upper right side of the stove next to the red object", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d361addd4233dde18166a4f.png"}, {"type": "image", "uri": "media/bdv2fail/aba1bcdd35abae2c307638e9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0093/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_107_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 93, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_107_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "remove the silver pan from the upper right side of stove", "subtask": "remove the silver pan from the upper right side of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2f947d4d4c049667755a07e7.png"}, {"type": "image", "uri": "media/bdv2fail/112623062501ef837e66e25a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0095/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 95, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the red spoon to the front right corner of the stove", "subtask": "move the red spoon to the front right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed39746680a92147fbfe4673.png"}, {"type": "image", "uri": "media/bdv2fail/612ffba806eb0336038c68fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0097/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 97, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the purple cloth to the upper left of the table", "subtask": "move the purple cloth to the upper left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/78281ccf49ecd2d72f02cc35.png"}, {"type": "image", "uri": "media/bdv2fail/78281ccf49ecd2d72f02cc35.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0100/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 100, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train", "episode_id": "12", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the silver pot on top of the toy mushroom.", "subtask": "move the silver pot on top of the toy mushroom.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dfd421c8d0c3cbba5e2cc99.png"}, {"type": "image", "uri": "media/bdv2fail/9c0371fec49a295951bdf16a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0101/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 101, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the metal pot to the left of the stove.", "subtask": "move the metal pot to the left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ce978411a6b19e94fbcaadd.png"}, {"type": "image", "uri": "media/bdv2fail/684f90ef6eaf3a4dfb791409.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0103/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 103, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the red spoon to the bottom left corner of the burner.", "subtask": "move the red spoon to the bottom left corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bfe26fc80a74a4ab7a70bf8b.png"}, {"type": "image", "uri": "media/bdv2fail/299d06fc30b1437f3087f8a8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0105/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_35_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 105, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_35_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the red spoon from the back left to the front right corner", "subtask": "move the red spoon from the back left to the front right corner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bffc7ba1b7cc212d6ed0a846.png"}, {"type": "image", "uri": "media/bdv2fail/ec02bd65b6583fd4c5d4b94a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0108/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 108, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "put the the fork to the top of the pot.", "subtask": "put the the fork to the top of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c83329436a8a36524b8ae752.png"}, {"type": "image", "uri": "media/bdv2fail/fd60c5307251f43e9854be6c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0109/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_43_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 109, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_43_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the yellow duck above the drawer", "subtask": "put the yellow duck above the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/41fcb48a11b60f5deb2b499d.png"}, {"type": "image", "uri": "media/bdv2fail/9e8f843e899beaa3131cc0e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0111/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/43", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 111, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "43", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the yellow towel beside the yellow microwave.", "subtask": "move the yellow towel beside the yellow microwave.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/99deef4da2a4180d5fbcc114.png"}, {"type": "image", "uri": "media/bdv2fail/333f6261cb1ce832a08d253b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0117/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 117, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the can to the lower left corner of the stove", "subtask": "move the can to the lower left corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/344616d784774ec413d342e1.png"}, {"type": "image", "uri": "media/bdv2fail/344616d784774ec413d342e1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0120/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 120, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "0", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the silver pot in the bottom right side of the burner", "subtask": "move the silver pot in the bottom right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0ddc5fe40decc5c40c65f500.png"}, {"type": "image", "uri": "media/bdv2fail/bc8ad945e340138b4bd66b25.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0123/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 123, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move silver pot to the top corner of the left side of the burner", "subtask": "move silver pot to the top corner of the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed600c58effec74a1024c79c.png"}, {"type": "image", "uri": "media/bdv2fail/88fd644504c0d1f258062405.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0126/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_70_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 126, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_70_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "move the yellow triangle from left to top of the table.", "subtask": "move the yellow triangle from left to top of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d0f2b998594da7d0f35bfba.png"}, {"type": "image", "uri": "media/bdv2fail/ec36e43732920ab5d43c21c4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0127/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_01_train/28", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 127, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_01_train", "episode_id": "28", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the cloth from right to left edge of the sink", "subtask": "move the cloth from right to left edge of the sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/175579bbb75198f80dce1115.png"}, {"type": "image", "uri": "media/bdv2fail/419b835930a7f158e21f582e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0128/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 128, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "pick up blue pen and put next to but not into drawer", "subtask": "pick up blue pen and put next to but not into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4233b866de77b685595cea8.png"}, {"type": "image", "uri": "media/bdv2fail/dbb82d5ec344c72ef5bdee42.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0129/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 129, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the red spoon to the upper right corner of the stove.", "subtask": "move the red spoon to the upper right corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/87ebcea7899966035986f523.png"}, {"type": "image", "uri": "media/bdv2fail/c47de527fc3df30fd301238a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0130/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_28_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 130, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_28_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the green object under the yellow towel.", "subtask": "put the green object under the yellow towel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9cbda08bed65dbad9149135.png"}, {"type": "image", "uri": "media/bdv2fail/fac9757433446bba7912b7f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0131/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_42_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 131, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_42_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "remove the blue spoon from the top of the green towel", "subtask": "remove the blue spoon from the top of the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5e256504f59abbfbc808e14.png"}, {"type": "image", "uri": "media/bdv2fail/b5e256504f59abbfbc808e14.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0132/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_31_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 132, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_31_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the green sponge to right of stove from left burner", "subtask": "move the green sponge to right of stove from left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0812e9a4727a17e46891ba5f.png"}, {"type": "image", "uri": "media/bdv2fail/7921bf8c20b4ffe262d88957.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0133/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 133, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put cucumber next to but not in cup", "subtask": "put cucumber next to but not in cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f5381e25124df9b002dcb3e1.png"}, {"type": "image", "uri": "media/bdv2fail/a88a871d37a7ec36a428f42e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0138/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/29", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 138, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "29", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the white object to the left burner", "subtask": "move the white object to the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5792dc5a2505cf8dbf06190.png"}, {"type": "image", "uri": "media/bdv2fail/c5792dc5a2505cf8dbf06190.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0139/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_15_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 139, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_15_train", "episode_id": "8", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "put the plastic ice on the bottom edge of the cloth", "subtask": "put the plastic ice on the bottom edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a7c05b89a50defa1b0ec5d34.png"}, {"type": "image", "uri": "media/bdv2fail/a7c05b89a50defa1b0ec5d34.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0140/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/18", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 140, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "18", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the blue strawberry from the right side of the table to the left.", "subtask": "move the blue strawberry from the right side of the table to the left.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c8dddd5afa781e2031e21f66.png"}, {"type": "image", "uri": "media/bdv2fail/360b11cc7d6327ef5326f027.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0142/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 142, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "put the blue spoon around but not on the top of the green towel", "subtask": "put the blue spoon around but not on the top of the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c728ac183e0042e2196d76b8.png"}, {"type": "image", "uri": "media/bdv2fail/b5e256504f59abbfbc808e14.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0144/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_31_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 144, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_31_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/513b8cad55480d76db8a44ed.png"}, {"type": "image", "uri": "media/bdv2fail/513b8cad55480d76db8a44ed.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0154/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 154, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train", "episode_id": "12", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "put the red fork on top of the grey plastic sink", "subtask": "put the red fork on top of the grey plastic sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79bbdba096a58685f6e196c3.png"}, {"type": "image", "uri": "media/bdv2fail/cc04978cf358f532cd05cd53.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0155/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_43_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 155, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_43_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the pot lid on the blue pot holder", "subtask": "move the pot lid on the blue pot holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/83233fc8f60c7caa1b1fa065.png"}, {"type": "image", "uri": "media/bdv2fail/07ec64235cdca9c08d37ba49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0156/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_11_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 156, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_11_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "put the orange cloth on top of the red object", "subtask": "put the orange cloth on top of the red object", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f7f2d6ef8aa1897c31a8dd0.png"}, {"type": "image", "uri": "media/bdv2fail/dfa272cabf1705b115676298.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0158/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 158, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the green spatula slighly to the left of the burner and place theblue thing below the burner", "subtask": "move the green spatula slighly to the left of the burner and place theblue thing below the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b474876d42b9d67e8e47bfbb.png"}, {"type": "image", "uri": "media/bdv2fail/18a30f73a9e0d9e6bef22fb1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0159/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 159, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 253, "role": "error_localization"}}
{"state": {"task": "move the cloth to the left of the pot.", "subtask": "move the cloth to the left of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e518077a57f79b81c7097538.png"}, {"type": "image", "uri": "media/bdv2fail/dc0da7265154d808b6b8fd75.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0160/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_06_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 160, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_06_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom right corner of the stove", "subtask": "move the silver pot to the bottom right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/82d3f4f7b498331b26b02b32.png"}, {"type": "image", "uri": "media/bdv2fail/125ed60fb0177a3bb7115edd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0161/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_47_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 161, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_47_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "place the red pepper close to but not on the blue towel", "subtask": "place the red pepper close to but not on the blue towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4063c0879df8e0e01c39798a.png"}, {"type": "image", "uri": "media/bdv2fail/8376c3ead33442a7aa79fd2f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0164/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_02_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 164, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_02_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the silver pot lid from the pot to the yellow cloth on the top right of the table", "subtask": "move the silver pot lid from the pot to the yellow cloth on the top right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8ccfe7756ea7cb41987b5ec2.png"}, {"type": "image", "uri": "media/bdv2fail/9b31e86b8bb31773b6d23303.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0166/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 166, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "put the yellow object next to but not on th ecloth in", "subtask": "put the yellow object next to but not on th ecloth in", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a48e1e2b8bbafae842bc5620.png"}, {"type": "image", "uri": "media/bdv2fail/18ce7a1e0faa7922f55bf051.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0167/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 167, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the green cloth to the bottom of the left burner", "subtask": "move the green cloth to the bottom of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2ea6339a920c791e06c9e5f2.png"}, {"type": "image", "uri": "media/bdv2fail/2ea6339a920c791e06c9e5f2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0168/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_40_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 168, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_40_train", "episode_id": "5", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "remove the grapes from the table and put it in the pan", "subtask": "remove the grapes from the table and put it in the pan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8bc3cfef84c83d95dc48e8c2.png"}, {"type": "image", "uri": "media/bdv2fail/8bc3cfef84c83d95dc48e8c2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0169/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_00_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 169, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_00_train", "episode_id": "3", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the cloth to the right of the toy sink", "subtask": "move the cloth to the right of the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e518077a57f79b81c7097538.png"}, {"type": "image", "uri": "media/bdv2fail/dc0da7265154d808b6b8fd75.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0174/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_06_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 174, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_06_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the yellow scrubber to the middle of the stove", "subtask": "move the yellow scrubber to the middle of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f36c8a64856c684950fe33dd.png"}, {"type": "image", "uri": "media/bdv2fail/46afe683a2a60d4e7e7b642e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0179/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 179, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put the ladle spoon on top of the plastic silver pot", "subtask": "put the ladle spoon on top of the plastic silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e7593a242e11348a1974bc1e.png"}, {"type": "image", "uri": "media/bdv2fail/f10af344b848619c590dd50f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0184/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_47_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 184, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_47_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "take the silver pot and put it underneath of the orange cloth", "subtask": "take the silver pot and put it underneath of the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/800698728a6ab05bbb50430a.png"}, {"type": "image", "uri": "media/bdv2fail/02c5858563adc22c6dddf39e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0187/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_40_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 187, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_40_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the napkin to the right of the pot and move the tomato towards it.", "subtask": "move the napkin to the right of the pot and move the tomato towards it.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c336591a6b66b41d921e61c7.png"}, {"type": "image", "uri": "media/bdv2fail/5e1cb8b87c1d92b68d150af2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0188/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_97_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 188, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_97_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "put banana around but not inside colander", "subtask": "put banana around but not inside colander", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac5998ec34d707221cff590a.png"}, {"type": "image", "uri": "media/bdv2fail/6edb1091c31cf813cc662ed3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0190/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train/32", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 190, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train", "episode_id": "32", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the blue spoon to the bottom right of the gray sink.", "subtask": "move the blue spoon to the bottom right of the gray sink.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/355752f3145a544ae2844058.png"}, {"type": "image", "uri": "media/bdv2fail/42e2bbe705ed69934072aa19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0193/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 193, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the corn on the lower left side of the right burner", "subtask": "put the corn on the lower left side of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3760b6926602a92f7e6e8c23.png"}, {"type": "image", "uri": "media/bdv2fail/7f4108fbe909082d4effca5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0194/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 194, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "slide the yellow spoon over to the front right corner of the stove.", "subtask": "slide the yellow spoon over to the front right corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2cc02ef2b86dd5c8959b69ac.png"}, {"type": "image", "uri": "media/bdv2fail/499bdb481bd0fa60997184ad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0195/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 195, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "remove the lid of the silver dish and put it on the upper right side of the table", "subtask": "remove the lid of the silver dish and put it on the upper right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e5e853f3b93796151fdc6af.png"}, {"type": "image", "uri": "media/bdv2fail/f7dc8eced90e464a81862ab2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0196/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_12_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 196, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_12_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 245, "role": "error_localization"}}
{"state": {"task": "move the wooden box to the drawer.", "subtask": "move the wooden box to the drawer.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc288b2bbc62dbcfc71c150c.png"}, {"type": "image", "uri": "media/bdv2fail/03550d729a91a62af8648858.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0197/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_26_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 197, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_26_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "moves green spatula to left side from the middle of stove", "subtask": "moves green spatula to left side from the middle of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b45f2201d5242af289fe6654.png"}, {"type": "image", "uri": "media/bdv2fail/5bc6786c9daed88027f82a1c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0198/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 198, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the green object next to the pot to upper right side of the stove", "subtask": "move the green object next to the pot to upper right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4aba18fc079ba1309e68b5a0.png"}, {"type": "image", "uri": "media/bdv2fail/9712fc022fdc49765ac64f4d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0199/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_86_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 199, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_86_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the blue spatula close by but not on the left burner", "subtask": "move the blue spatula close by but not on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/783b3ec5d8c46e163ee8e7c4.png"}, {"type": "image", "uri": "media/bdv2fail/b24e1d8d09e7f4325e6abbf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0201/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 201, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the silver bowl to the top left side of the table.", "subtask": "move the silver bowl to the top left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ac07bb9fabdcee6bc274b0f.png"}, {"type": "image", "uri": "media/bdv2fail/3cf39080b46015547b130b17.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0202/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 202, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the pear to the bottom right of the burner", "subtask": "move the pear to the bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/31e659ab4e8709fe59fd0106.png"}, {"type": "image", "uri": "media/bdv2fail/f291024a02fa39a39920610e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0204/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 204, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "take the yellow brush and put it on the lid", "subtask": "take the yellow brush and put it on the lid", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/02a037815b1c3049a02b788e.png"}, {"type": "image", "uri": "media/bdv2fail/8bc97a917bcc57360dd525b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0206/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/18", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 206, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put blueberries close by but not on plate sink", "subtask": "put blueberries close by but not on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/59c7373c6e708915ebcd6c37.png"}, {"type": "image", "uri": "media/bdv2fail/cb2ebef068c027d40732b148.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0209/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 209, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the blue spoon to the right side of the silver pot from the center of the induction stove.", "subtask": "move the blue spoon to the right side of the silver pot from the center of the induction stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd4a34150bbda26b93e0d803.png"}, {"type": "image", "uri": "media/bdv2fail/3cd21797217244e8c8f8aa71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0211/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 211, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "move the toy microwave to the lower middle part of the table.", "subtask": "move the toy microwave to the lower middle part of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d31a1dd75842175c6d418392.png"}, {"type": "image", "uri": "media/bdv2fail/c13eb30d7d3bf39867b9de12.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0212/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 212, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the silver pot from the top left corner of the stove to the stove", "subtask": "move the silver pot from the top left corner of the stove to the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/051f24ca4e6f0d672d3560ca.png"}, {"type": "image", "uri": "media/bdv2fail/aca3a0b630d2b7fd14bfef3e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0214/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 214, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "close a small box", "subtask": "close a small box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2418d9e60ad3349e6fe07d52.png"}, {"type": "image", "uri": "media/bdv2fail/73c942b2dcc8550c26354fa9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0215/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 215, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "move the blue toothbrush to the right side of the burner.", "subtask": "move the blue toothbrush to the right side of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/998a936ac6f2554536887c4c.png"}, {"type": "image", "uri": "media/bdv2fail/243b022a618e6d43f2dd5efb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0218/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_03_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 218, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_03_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "moves the rubber chicken from upper left corner to bottom left corner", "subtask": "moves the rubber chicken from upper left corner to bottom left corner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9843fb458a76605ed1c93ff6.png"}, {"type": "image", "uri": "media/bdv2fail/2c36533e7ee2ac3ab2fc4107.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0226/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 226, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the cloth to the bottom right corner of the burner", "subtask": "move the cloth to the bottom right corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3f1e7392f8ded1be8e1c0536.png"}, {"type": "image", "uri": "media/bdv2fail/b4646bb32768c7ad9c74cc42.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0228/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 228, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the silver spoon to the middle side of the table.", "subtask": "move the silver spoon to the middle side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b84674213e777dc0fd1805c1.png"}, {"type": "image", "uri": "media/bdv2fail/dc837253f18a1bc33bf26deb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0229/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 229, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the grey sink", "subtask": "move the silver pot to the grey sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/18d9ae62293640cb5a8c95b9.png"}, {"type": "image", "uri": "media/bdv2fail/aa863f51b10dddb658775fba.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0231/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_39_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 231, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_39_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "pick up blue screwdriver.", "subtask": "pick up blue screwdriver.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/93420922681c8ce81cef8b98.png"}, {"type": "image", "uri": "media/bdv2fail/cba28588f6ac3c325745016e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0232/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train/40", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 232, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train", "episode_id": "40", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "error_localization"}}
{"state": {"task": "put the blue spatula out of the pan.", "subtask": "put the blue spatula out of the pan.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e6414471fd3fc9bb921d3fd.png"}, {"type": "image", "uri": "media/bdv2fail/369936818ed3f87c451d3ac2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0234/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 234, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the grey cloth to the bottom middle part of the stove.", "subtask": "move the grey cloth to the bottom middle part of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f4d612bda78736c843f1516.png"}, {"type": "image", "uri": "media/bdv2fail/67e0e7a8f1705208c81cd2de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0235/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_09_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 235, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_09_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "take the silver pot and cook it with a red pepper", "subtask": "take the silver pot and cook it with a red pepper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/77f9aeca510431f1129b3219.png"}, {"type": "image", "uri": "media/bdv2fail/6cfcb102525afcdaf5cbc00e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0236/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 236, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put the blue spoon on top of the green cloth.", "subtask": "put the blue spoon on top of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/888867d29ffd0d9ab5038f7b.png"}, {"type": "image", "uri": "media/bdv2fail/63fcd9930f4272aa1004534d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0237/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 237, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the left burner.", "subtask": "move the silver pot to the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f2a47d69693419ed1d256563.png"}, {"type": "image", "uri": "media/bdv2fail/dba612923759fb5178cb07de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0238/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 238, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c7d988fe071fb8d1d4eba40.png"}, {"type": "image", "uri": "media/bdv2fail/7c7d988fe071fb8d1d4eba40.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0239/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 239, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 215, "role": "error_localization"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bce3ef38009e6caccd0c94a2.png"}, {"type": "image", "uri": "media/bdv2fail/3d29fef4bfce5020e820eeb6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0242/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_13_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 242, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_13_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 215, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from left to right", "subtask": "unfold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/59123a67f89cc4fb302d117f.png"}, {"type": "image", "uri": "media/bdv2fail/59123a67f89cc4fb302d117f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0244/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 244, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train", "episode_id": "5", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "put yellow toy banana in blue plate", "subtask": "put yellow toy banana in blue plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eb9313a0269ac2ad9313717e.png"}, {"type": "image", "uri": "media/bdv2fail/4cd407928c4523cb7804dcd6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0247/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train/39", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 247, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train", "episode_id": "39", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "put toy lemon from counter to sink", "subtask": "put toy lemon from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c85b8a4fa2b7f7ac499c9a1e.png"}, {"type": "image", "uri": "media/bdv2fail/8e871c8d4ae3e86b344448ec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0248/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/26", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 248, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "26", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "place the yellow banana to the top left corner of the plastic sink", "subtask": "place the yellow banana to the top left corner of the plastic sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70a8029ed3e32a43df6163e3.png"}, {"type": "image", "uri": "media/bdv2fail/9348e78412fd0a85d4bdebe3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0249/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 249, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "close the lip of silver pot and move to the right upper side", "subtask": "close the lip of silver pot and move to the right upper side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/21e8524b6b602ef4ee0ea89b.png"}, {"type": "image", "uri": "media/bdv2fail/b71a6c4d6d10c4a00f6c8aa8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0252/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 252, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the silver pan to the right side of the stove", "subtask": "move the silver pan to the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/060fb26aa869ed92663da79c.png"}, {"type": "image", "uri": "media/bdv2fail/19cb7c8614612be205f3df82.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0253/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_41_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 253, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_41_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the blue spoon and sushi to the right side of the table.", "subtask": "move the blue spoon and sushi to the right side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/98b9ab16e1120a31bab9cc59.png"}, {"type": "image", "uri": "media/bdv2fail/1b330d779384f39d4fd708a5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0254/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_01_train/18", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 254, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_01_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the red toy fork to the right side of the pot.", "subtask": "move the red toy fork to the right side of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9aec55453901a208bc241d7.png"}, {"type": "image", "uri": "media/bdv2fail/9ac3ac418bdd7d8694d61b54.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0257/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_07_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 257, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_07_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the red spoon in the middle and the left side of silver pot", "subtask": "move the red spoon in the middle and the left side of silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2449f33c97b6a9259171a677.png"}, {"type": "image", "uri": "media/bdv2fail/7baebfa4bb6e75cb389e3547.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0258/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_17_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 258, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_17_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "put the silver lid on the green toy spoon.", "subtask": "put the silver lid on the green toy spoon.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/12a3982a84be10dcc2289c44.png"}, {"type": "image", "uri": "media/bdv2fail/be32103799c792ede7b439c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0260/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 260, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "take the silver lid off the silver pot and put it on the right side of the silver pot", "subtask": "take the silver lid off the silver pot and put it on the right side of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e07266de493f89fdd46818e.png"}, {"type": "image", "uri": "media/bdv2fail/c07720c11cb973ed6a9ed9a8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0265/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 265, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "put the blue sponge on the upper right corner of the burner.", "subtask": "put the blue sponge on the upper right corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/858d719c2d354107c5d56c7a.png"}, {"type": "image", "uri": "media/bdv2fail/0a44be1c6ddf79debf9727a2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0266/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 266, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put carrot beside but not on plate", "subtask": "put carrot beside but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79a7f3e36889ede4416f5487.png"}, {"type": "image", "uri": "media/bdv2fail/681efcafab9b969741ab6378.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0267/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_carrot_on_plate_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 267, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_carrot_on_plate_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "remove the cloth from the bottom left corner of the table", "subtask": "remove the cloth from the bottom left corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c9f36085fe5ebe76674c6d70.png"}, {"type": "image", "uri": "media/bdv2fail/1d71cf0ef09e96a354cf5d57.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0268/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_33_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 268, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_33_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the silver pot on the right side of stove", "subtask": "move the silver pot on the right side of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bfbb17c2d31319e7c2ca91ba.png"}, {"type": "image", "uri": "media/bdv2fail/113a177f72f1c467c87d34ab.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0269/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_60_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 269, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_60_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the grey plastic to the right burner.", "subtask": "move the grey plastic to the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/65d4df1fed899e90ea46c903.png"}, {"type": "image", "uri": "media/bdv2fail/ba7acda5635efff78f9133c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0271/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_33_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 271, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_33_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "take the silver vessel and place it close to but not on the purple cloth.", "subtask": "take the silver vessel and place it close to but not on the purple cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/62f99b51f3c8b8557574d4a6.png"}, {"type": "image", "uri": "media/bdv2fail/e4acfcdcd6ae2ce11e376c73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0274/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_02_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 274, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_02_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the pot to the left", "subtask": "move the pot to the left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ffc9d29910b039505829b301.png"}, {"type": "image", "uri": "media/bdv2fail/76ad96f7b228ea3c988eaf84.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0275/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_40_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 275, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_40_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "move the blue spatula to the right side of the burner", "subtask": "move the blue spatula to the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14f4a698e896767d3f97c4b7.png"}, {"type": "image", "uri": "media/bdv2fail/7fd933337a022e5ea68b5f0c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0276/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 276, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the blue carpet to left from right of the table.", "subtask": "move the blue carpet to left from right of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2d0e2c1c80a630c58f265efd.png"}, {"type": "image", "uri": "media/bdv2fail/951bcbca10bfb008825032bd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0278/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 278, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the yellow object from the pot to the right of the left side burner.", "subtask": "move the yellow object from the pot to the right of the left side burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0e41312cca45a06022e86177.png"}, {"type": "image", "uri": "media/bdv2fail/8eff56aaacb09f759a791e04.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0279/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 279, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the can to the left side of the stove", "subtask": "move the can to the left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc01f813c13beb6cbcd794e6.png"}, {"type": "image", "uri": "media/bdv2fail/43a134743df93288cc3e3c78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0280/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 280, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the gray bowl to the right of the burner.", "subtask": "move the gray bowl to the right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/002a2ed005f8663642682f2c.png"}, {"type": "image", "uri": "media/bdv2fail/61e44c16422e67a3b8960ece.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0281/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_60_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 281, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_60_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "take the silver pot and put it on the top of the stove.", "subtask": "take the silver pot and put it on the top of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/838447fa7165c95c96582620.png"}, {"type": "image", "uri": "media/bdv2fail/cd18fcf7ee91946939bdeb3c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0284/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 284, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "remove the lid from the metal pot and put it underneath of the green cloth", "subtask": "remove the lid from the metal pot and put it underneath of the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/de9396e1269d42f4ff061a1c.png"}, {"type": "image", "uri": "media/bdv2fail/b7a77f5c541e9cbdd257349b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0285/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 285, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "place the red fork on the stove.", "subtask": "place the red fork on the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a4b0da750d8bc57c7bc089e6.png"}, {"type": "image", "uri": "media/bdv2fail/42889ccf9735c917738a94e5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0289/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_70_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 289, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_70_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "put the green object close by but not in between the burners", "subtask": "put the green object close by but not in between the burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4585d0acc38272f46e71d7c4.png"}, {"type": "image", "uri": "media/bdv2fail/6ea37b21854ef0c98071472d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0291/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 291, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the black tape to top from bottom", "subtask": "move the black tape to top from bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fbd5256059f152bb7404ce4b.png"}, {"type": "image", "uri": "media/bdv2fail/72d30aa3d34fb9a81eedb5e2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0292/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 292, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the blue fork to the bottom right corner of the toy sink", "subtask": "move the blue fork to the bottom right corner of the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c959092d6217d1397748c032.png"}, {"type": "image", "uri": "media/bdv2fail/96784e8df9e2a8d2c2c4b2de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0295/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 295, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "put the green object next to but not in the silver pot", "subtask": "put the green object next to but not in the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a579474f3d1b86375cbf1b68.png"}, {"type": "image", "uri": "media/bdv2fail/f5009d123f10282ba5b8ac73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0298/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_74_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 298, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_74_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "remove the blue fork from the upper left corner of the burner", "subtask": "remove the blue fork from the upper left corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a44be1c6ddf79debf9727a2.png"}, {"type": "image", "uri": "media/bdv2fail/858d719c2d354107c5d56c7a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0299/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 299, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the toy microwave to the bottom of the table.", "subtask": "move the toy microwave to the bottom of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a628c2c636d373ca727d8859.png"}, {"type": "image", "uri": "media/bdv2fail/ffdce5de74188e5b350235ec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0300/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 300, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "fold the cloth from the right to the left", "subtask": "fold the cloth from the right to the left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/17474648a31be063f2779a23.png"}, {"type": "image", "uri": "media/bdv2fail/17474648a31be063f2779a23.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0301/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 301, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "21", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the yellow plastic banana to the top left corner of the burner.", "subtask": "move the yellow plastic banana to the top left corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a95af49e43a45c5edb6c465e.png"}, {"type": "image", "uri": "media/bdv2fail/c051ac363eda61529c203fdf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0302/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 302, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "open low fridge", "subtask": "open low fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/821d46f9e40ba9b079289724.png"}, {"type": "image", "uri": "media/bdv2fail/e3f6a6e85b22a52744e0471c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0303/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen5_close_low_fridge_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 303, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen5_close_low_fridge_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 215, "role": "error_localization"}}
{"state": {"task": "move the blue pot holder to the bottom left burner.", "subtask": "move the blue pot holder to the bottom left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d3c0923983210a72c9e8a47.png"}, {"type": "image", "uri": "media/bdv2fail/17f02f73338b9457674f19d4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0305/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_29_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 305, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_29_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the green toy to the left corner of the burner.", "subtask": "move the green toy to the left corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a9ba20213047b09ee58eb5fb.png"}, {"type": "image", "uri": "media/bdv2fail/7440e177733e26251e68c520.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0306/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_83_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 306, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_83_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put the lid on the grey pot", "subtask": "put the lid on the grey pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/34d0128fad3c66f26c4e8fe9.png"}, {"type": "image", "uri": "media/bdv2fail/d39bcfec2d4b3e701c515546.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0307/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 307, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the blue spoon between the silver pot and the mushroom", "subtask": "move the blue spoon between the silver pot and the mushroom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86d87d5c3500dd85230c7cf5.png"}, {"type": "image", "uri": "media/bdv2fail/adf8de94b21e92fae9a3e85e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0308/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_95_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 308, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_95_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the pot above the faucet.", "subtask": "move the pot above the faucet.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/789dd36199509b6650a94861.png"}, {"type": "image", "uri": "media/bdv2fail/6dd027b67403e580ccc0cd19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0310/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_64_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 310, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_64_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "take the small blue spoon to the toy stove", "subtask": "take the small blue spoon to the toy stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e0e035ee5fd5dc53cc69789b.png"}, {"type": "image", "uri": "media/bdv2fail/b070e974963fa86b9b473135.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0312/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_110_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 312, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_110_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "picking up the pot and putting it on the yellow sponge", "subtask": "picking up the pot and putting it on the yellow sponge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eefd2a61d33f23500a58a79e.png"}, {"type": "image", "uri": "media/bdv2fail/d16b960af03243034ee49a8f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0313/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_99_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 313, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_99_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "remove cup from plate", "subtask": "remove cup from plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a7a4e8599971151e571b8b49.png"}, {"type": "image", "uri": "media/bdv2fail/a7a4e8599971151e571b8b49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0314/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/23", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 314, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "23", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "put the silver pan on the upper right side of stove", "subtask": "put the silver pan on the upper right side of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/112623062501ef837e66e25a.png"}, {"type": "image", "uri": "media/bdv2fail/2f947d4d4c049667755a07e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0315/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 315, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the red pepper to the upper left side of the table.", "subtask": "move the red pepper to the upper left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/35e4f540ff8783bb14a295ab.png"}, {"type": "image", "uri": "media/bdv2fail/6721af84bd7556d1bed8aa59.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0316/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 316, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the yellow knife to the left side of the burner", "subtask": "move the yellow knife to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c1fb4f537cbe91532a67feb.png"}, {"type": "image", "uri": "media/bdv2fail/1c1fb4f537cbe91532a67feb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0318/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 318, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put the green eggplant on the white tile backsplash", "subtask": "put the green eggplant on the white tile backsplash", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e71541678506534af88b5816.png"}, {"type": "image", "uri": "media/bdv2fail/26d4a79f77d1865e2e5cf3e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0319/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_42_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 319, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_42_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the pot below the right burner", "subtask": "move the pot below the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/789dd36199509b6650a94861.png"}, {"type": "image", "uri": "media/bdv2fail/6dd027b67403e580ccc0cd19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0320/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_64_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 320, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_64_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the green toy to the top edge of the cloth from the left of the table.", "subtask": "move the green toy to the top edge of the cloth from the left of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e8f418c4d9bdf0fc7f0a5305.png"}, {"type": "image", "uri": "media/bdv2fail/399c74ecbdf93fd99c244ccb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0325/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/24", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 325, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "24", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "remove the silver pot from the yellow cloth and place it in the lower right corner of the table", "subtask": "remove the silver pot from the yellow cloth and place it in the lower right corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/26b250a85408ec994349a45a.png"}, {"type": "image", "uri": "media/bdv2fail/a86bfea41967df9f35b88d61.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0328/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 328, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "put the blue spoon to the center of the burner.", "subtask": "put the blue spoon to the center of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a11e82fa3b7b4b4505aa0993.png"}, {"type": "image", "uri": "media/bdv2fail/d2ba3f4d0d1d4c85b9914e41.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0330/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 330, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the yellow object to the lower right side of the desk", "subtask": "move the yellow object to the lower right side of the desk", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/00382afac86ab19d782ef023.png"}, {"type": "image", "uri": "media/bdv2fail/00382afac86ab19d782ef023.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0331/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 331, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the orange towel from the top left to the centre of the table.", "subtask": "move the orange towel from the top left to the centre of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/48eab26340f105a65c972f11.png"}, {"type": "image", "uri": "media/bdv2fail/b919347f16bf703468eb7c40.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0337/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_17_train/22", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 337, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_17_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "take the orange spoon to the left side of the burner", "subtask": "take the orange spoon to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a4f5aa85aa9812d84d9c123f.png"}, {"type": "image", "uri": "media/bdv2fail/956140b49fb2a3b519957bbe.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0338/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_17_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 338, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_17_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "place the blue plastic toy on the front left edge of the stove", "subtask": "place the blue plastic toy on the front left edge of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a49ff3162c5c99ba207f9477.png"}, {"type": "image", "uri": "media/bdv2fail/9158e2ea2fd8b8337b56cea7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0341/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_30_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 341, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_30_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the blue spoon to the bottom corner of the side of the pot.", "subtask": "move the blue spoon to the bottom corner of the side of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ae91a1619af58fbb83ab210c.png"}, {"type": "image", "uri": "media/bdv2fail/f1e2447f5e1c368f1584c44c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0342/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 342, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "put the blue cup into the yellow basket", "subtask": "put the blue cup into the yellow basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b321bf8039405b2c9c4ac518.png"}, {"type": "image", "uri": "media/bdv2fail/fb54d56184efbb83e25a2d23.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0344/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/67", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 344, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "67", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move blue doll from top of the drawer and place it close by but not on the table", "subtask": "move blue doll from top of the drawer and place it close by but not on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ad6671a2052a36849b18cad.png"}, {"type": "image", "uri": "media/bdv2fail/65e4ea24f5339e34c0d3136c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0345/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 345, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 245, "role": "error_localization"}}
{"state": {"task": "take the lid and put it on the stove", "subtask": "take the lid and put it on the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8d38b21457bab8bf300762b.png"}, {"type": "image", "uri": "media/bdv2fail/2541d96eb31e2960043b6478.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0346/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 346, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the cloth to the middle from the top edge of the table.", "subtask": "move the cloth to the middle from the top edge of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c52c5f804f16982ef87d98b9.png"}, {"type": "image", "uri": "media/bdv2fail/f98fa713ba1b7dfdeb87df89.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0347/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 347, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom left corner of the burners.", "subtask": "move the silver pot to the bottom left corner of the burners.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/73cf353cd0a0d5851fdd2963.png"}, {"type": "image", "uri": "media/bdv2fail/51928347ebfc3b06eab13676.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0349/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 349, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the spatula in the upper left corner of the table.", "subtask": "move the spatula in the upper left corner of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e49513dc0dc844d942fd119f.png"}, {"type": "image", "uri": "media/bdv2fail/91f8105116360b7632a8ee77.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0350/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_02_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 350, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_02_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from top left to left", "subtask": "unfold the cloth from top left to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f1bfee9fd3421583484942e5.png"}, {"type": "image", "uri": "media/bdv2fail/f1bfee9fd3421583484942e5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0351/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_07_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 351, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_07_train", "episode_id": "11", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the upper left corner of the stove.", "subtask": "move the blue cloth to the upper left corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c480b856b155bdef3d96843.png"}, {"type": "image", "uri": "media/bdv2fail/24913f5b7db5a48dce285510.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0353/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_93_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 353, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_93_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "remove spatula from cutting board", "subtask": "remove spatula from cutting board", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/51ac62691acd8fce98355fc9.png"}, {"type": "image", "uri": "media/bdv2fail/51ac62691acd8fce98355fc9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0354/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_spatula_on_cutting_board_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 354, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_spatula_on_cutting_board_train", "episode_id": "2", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "place the spoon at the right side of the stove.", "subtask": "place the spoon at the right side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8d3824123baf606e3bbcc846.png"}, {"type": "image", "uri": "media/bdv2fail/0f5154b8f3c8ba368e0397e4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0356/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_35_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 356, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_35_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth to the bottom right of the burner", "subtask": "move the yellow cloth to the bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/112df704260b5e2cb3057c14.png"}, {"type": "image", "uri": "media/bdv2fail/eb200e445e07c6be11b99dd4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0358/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 358, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the plastic toy egg to the bottom left of the stove.", "subtask": "move the plastic toy egg to the bottom left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/983f41bd20a6a51e416e8cd0.png"}, {"type": "image", "uri": "media/bdv2fail/05d7c516999e114f886ad3fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0361/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 361, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth to the top of the ice cream cone", "subtask": "move the yellow cloth to the top of the ice cream cone", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6587edd0e533db315beeedfb.png"}, {"type": "image", "uri": "media/bdv2fail/0043486c552bff5e13c89b03.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0363/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 363, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the green monkey toy to the middle bottom of the drawer", "subtask": "move the green monkey toy to the middle bottom of the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c229c1ccf96ee54df909ea64.png"}, {"type": "image", "uri": "media/bdv2fail/2b71a0a9260801b6748975d3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0364/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 364, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "take the toy sink and place it on top of the yellow cloth.", "subtask": "take the toy sink and place it on top of the yellow cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d9fa9b580af0720c76fe6ae0.png"}, {"type": "image", "uri": "media/bdv2fail/01f295430c8289188483bb1e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0367/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 367, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "put the plastic pink cup on the yellow dish drying rack.", "subtask": "put the plastic pink cup on the yellow dish drying rack.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c583f095c9071d4e3bca7aa.png"}, {"type": "image", "uri": "media/bdv2fail/ed8be19109dc9a2b044da7aa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0369/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 369, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the yellow object below the right burner and close to spoon.", "subtask": "move the yellow object below the right burner and close to spoon.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5406d3fac0e9ca218c4d4098.png"}, {"type": "image", "uri": "media/bdv2fail/57d35c8e862952f255dfb1e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0371/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_48_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 371, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_48_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the green cloth below the silver pan on the bottom left of burner", "subtask": "move the green cloth below the silver pan on the bottom left of burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/487b70f81eccf9f869339f1e.png"}, {"type": "image", "uri": "media/bdv2fail/c47647d7a728c6b4c6dc554b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0373/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 373, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "move the blue dishrag to the right side of the burner", "subtask": "move the blue dishrag to the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9a6418fb9aada573bf773f2.png"}, {"type": "image", "uri": "media/bdv2fail/bfd5bcd54254f2b401a88dbf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0374/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 374, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the violet cloth in the bottom right corner of the toy sink", "subtask": "put the violet cloth in the bottom right corner of the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ecf2cf92e22bf08d966291b0.png"}, {"type": "image", "uri": "media/bdv2fail/1463de8f48b1d9c93229d0de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0376/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_84_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 376, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_84_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the orange toy to the left side of the table from the middle upper part", "subtask": "move the orange toy to the left side of the table from the middle upper part", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}, {"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0378/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 378, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the upperleft of table", "subtask": "move the silver pot to the upperleft of table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/411c69c79a0e1233cc485d4c.png"}, {"type": "image", "uri": "media/bdv2fail/411c69c79a0e1233cc485d4c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0379/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 379, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "remove the green vegetable from the right side of the burner", "subtask": "remove the green vegetable from the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec5cb938ff4a1a745a46f539.png"}, {"type": "image", "uri": "media/bdv2fail/ec5cb938ff4a1a745a46f539.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0382/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 382, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train", "episode_id": "11", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put the red fork close by but not on the green cloth", "subtask": "put the red fork close by but not on the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/42bd23e905406794665e8927.png"}, {"type": "image", "uri": "media/bdv2fail/70837f42c84534f5ea69ad69.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0386/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 386, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "cover the silver pot and put the lid on the left side of the table", "subtask": "cover the silver pot and put the lid on the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9288175aa9efc795575e0052.png"}, {"type": "image", "uri": "media/bdv2fail/f814cf96c5dcf1a5956d47b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0387/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 387, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the blue towel from the left burner to the right burner.", "subtask": "move the blue towel from the left burner to the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0475b18118c518a6e66c64b7.png"}, {"type": "image", "uri": "media/bdv2fail/e9c678dba3c0355a6ece7fa3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0388/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 388, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the bread to the top of the left burner", "subtask": "move the bread to the top of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d8fe49c3e772f22e6f6be79c.png"}, {"type": "image", "uri": "media/bdv2fail/d45104f99887632d068fbdbb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0392/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 392, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the orange toy from the right side of the table to the middle upper part", "subtask": "move the orange toy from the right side of the table to the middle upper part", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/92375aa847b78ec7040b05b6.png"}, {"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0394/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 394, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "remove cup from plate", "subtask": "remove cup from plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6b9f1ba9d9e66b13306a4f11.png"}, {"type": "image", "uri": "media/bdv2fail/6b9f1ba9d9e66b13306a4f11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0395/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/27", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 395, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "27", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "put the silver pot between the stove and the blue sponge", "subtask": "put the silver pot between the stove and the blue sponge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1290672c526271caffc13fcd.png"}, {"type": "image", "uri": "media/bdv2fail/dbd9643f67868490ac611e53.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0397/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_15_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 397, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_15_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the small silver pot to the lower left corner of the stove.", "subtask": "move the small silver pot to the lower left corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/344616d784774ec413d342e1.png"}, {"type": "image", "uri": "media/bdv2fail/50035b9dd3d241c717dad202.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0398/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 398, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "0", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the blue brush to the upper right of the green sponge.", "subtask": "move the blue brush to the upper right of the green sponge.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/24887e88fd1ff5baede16f64.png"}, {"type": "image", "uri": "media/bdv2fail/e1172e7d277e9ba13b760c50.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0399/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_61_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 399, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_61_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the red fork to the left side of the burner", "subtask": "move the red fork to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/69e7c3d68d65043fd2ebfa4a.png"}, {"type": "image", "uri": "media/bdv2fail/baa8a599132a52b38d993198.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0400/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 400, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "moves the silver lid to the left side of a green cloth from the upper side of the table, tin", "subtask": "moves the silver lid to the left side of a green cloth from the upper side of the table, tin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d6dfa1d37a2b0d7c2c0d59b.png"}, {"type": "image", "uri": "media/bdv2fail/572fff386ad7b684187377a9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0403/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 403, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 251, "role": "error_localization"}}
{"state": {"task": "move the toy cheese into the toy sink", "subtask": "move the toy cheese into the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dda3c89983bad438cea10f4d.png"}, {"type": "image", "uri": "media/bdv2fail/413372a83226654b018cea8e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0404/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 404, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6c28b89efe20cd79957ea34.png"}, {"type": "image", "uri": "media/bdv2fail/bf7cdaf9abe2e267c08b44eb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0405/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_13_train/38", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 405, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_13_train", "episode_id": "38", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 215, "role": "error_localization"}}
{"state": {"task": "put the can of soup at the bottom of the left burner", "subtask": "put the can of soup at the bottom of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3ade824fdf797412725045a9.png"}, {"type": "image", "uri": "media/bdv2fail/3d625433c1ca074e05da8fee.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0407/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 407, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the yellow duck on the drawer", "subtask": "put the yellow duck on the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d4f4aad820e54238a4d4bf87.png"}, {"type": "image", "uri": "media/bdv2fail/cadf3573d63e2515c5e0acd8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0412/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/30", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 412, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "30", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the lemon on the left side of the stove.", "subtask": "move the lemon on the left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6298e643302c41acc33c2f7d.png"}, {"type": "image", "uri": "media/bdv2fail/fc337b352444ee9b6bf7ffd8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0413/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_24_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 413, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_24_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the blue brush to the top left of the stove", "subtask": "move the blue brush to the top left of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/47febdcd44b00f869f4a49dc.png"}, {"type": "image", "uri": "media/bdv2fail/b8af5ecdad8ad1771ef5311a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0414/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 414, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "moves the ribbon to right of the cloth from top.", "subtask": "moves the ribbon to right of the cloth from top.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4ebbc64e9b733f0257e30a7e.png"}, {"type": "image", "uri": "media/bdv2fail/43eea7b49ab6e0a76025341e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0416/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/34", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 416, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "34", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the red pepper around but not inside the pot", "subtask": "put the red pepper around but not inside the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/58674116eedb2746df95954d.png"}, {"type": "image", "uri": "media/bdv2fail/eed54a2f825d6c9b9913a180.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0417/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_90_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 417, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_90_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the pot close to but not on the napkin.", "subtask": "put the pot close to but not on the napkin.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2d84cf6c2cdd10010e8845ca.png"}, {"type": "image", "uri": "media/bdv2fail/d55d5b84b83ab3e4a0c2ddc7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0418/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_01_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 418, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_01_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "take the silver pot to the left side of the left burner", "subtask": "take the silver pot to the left side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/84a7a0fa0c9d8193ab4633a9.png"}, {"type": "image", "uri": "media/bdv2fail/bcad9b238f6ede573da8bc0a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0420/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 420, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the yellow spoon to the bottom part between the two burners.", "subtask": "move the yellow spoon to the bottom part between the two burners.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c1298032a108bbb4f814c176.png"}, {"type": "image", "uri": "media/bdv2fail/714b479e2a29064bab0aa51f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0421/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_108_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 421, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_108_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "take the banana and put it on the right side of the table", "subtask": "take the banana and put it on the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/57fae9750910a27467472768.png"}, {"type": "image", "uri": "media/bdv2fail/cc562e0ad1c2faf55838acc6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0422/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train/18", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 422, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "turn toy banana front to right.", "subtask": "turn toy banana front to right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6fb5a93e0128640e0b986cc1.png"}, {"type": "image", "uri": "media/bdv2fail/ee00a5a7595de6be9cd6df22.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0423/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train/19", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 423, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train", "episode_id": "19", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "move the orange cloth from the left to the grey sink", "subtask": "move the orange cloth from the left to the grey sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/074338752e9d350bb97a6c61.png"}, {"type": "image", "uri": "media/bdv2fail/3337c2945481b99ef0c3388d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0426/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_25_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 426, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_25_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put the silver pot on the red and yellow knobs nearby", "subtask": "put the silver pot on the red and yellow knobs nearby", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14c5a0d430c195739fc480ee.png"}, {"type": "image", "uri": "media/bdv2fail/e9b3e7d0d0e1360d48d4bbc3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0428/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_06_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 428, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_06_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "pick up the red toolbox", "subtask": "pick up the red toolbox", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/020cbd64b344ff628c55c6a1.png"}, {"type": "image", "uri": "media/bdv2fail/5810bcfb436950bfe2e244c7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0432/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_violet_Allen_key_train/27", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 432, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_violet_Allen_key_train", "episode_id": "27", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "error_localization"}}
{"state": {"task": "move the red knobs to the bottom right of the burner.", "subtask": "move the red knobs to the bottom right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/33970f2384e0ff09de0b4e85.png"}, {"type": "image", "uri": "media/bdv2fail/721d977520e27aa1272bf3b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0434/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_11_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 434, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_11_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "place tomato inside towel", "subtask": "place tomato inside towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4811e7b50ac3b7897405ed16.png"}, {"type": "image", "uri": "media/bdv2fail/a2f85553fc00378500d441fc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0437/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_97_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 437, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_97_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "move the sponge above the left burner.", "subtask": "move the sponge above the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/75ac8805e4280241179870cd.png"}, {"type": "image", "uri": "media/bdv2fail/c644e8e7429bf5ca2c5c9cfd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0438/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_75_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 438, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_75_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the blue plastic item to the corner near the vessel.", "subtask": "move the blue plastic item to the corner near the vessel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91b29402a11de413e7d29c07.png"}, {"type": "image", "uri": "media/bdv2fail/56f181de99b94497aab18c27.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0439/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_03_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 439, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_03_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom of the green cloth.", "subtask": "move the silver pot to the bottom of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8339c0e465a9ce5e768cde06.png"}, {"type": "image", "uri": "media/bdv2fail/058e5b0114c6c0797faa1b07.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0442/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 442, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the cloth to the left of the metal tool.", "subtask": "move the cloth to the left of the metal tool.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/20030bbeac881a51e0322251.png"}, {"type": "image", "uri": "media/bdv2fail/54d4853d62d68aca90950647.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0443/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_04_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 443, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_04_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the blue fork underneath of the orange cloth", "subtask": "move the blue fork underneath of the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4cab7430981ffa604610e270.png"}, {"type": "image", "uri": "media/bdv2fail/316fdb7ecf16c7a7ebd0847d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0444/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_97_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 444, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_97_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from top to bottom", "subtask": "unfold the cloth from top to bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e8d74c530d63a73e98553491.png"}, {"type": "image", "uri": "media/bdv2fail/e8d74c530d63a73e98553491.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0447/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_03_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 447, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_03_train", "episode_id": "1", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "place the squash inside the stove", "subtask": "place the squash inside the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4c13c10f869e711c48357af2.png"}, {"type": "image", "uri": "media/bdv2fail/0bebd2a16b3c27a3f429f688.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0450/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_33_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 450, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_33_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "move the ladle around but not on the upper side of the table", "subtask": "move the ladle around but not on the upper side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/37cf229759e286a12b595942.png"}, {"type": "image", "uri": "media/bdv2fail/95aa020ef71ae1fb299009d3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0451/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_06_train/20", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 451, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_06_train", "episode_id": "20", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "put the yellow scrubber near but not on the green cloth.", "subtask": "put the yellow scrubber near but not on the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/def3e801c49976c0d1f75979.png"}, {"type": "image", "uri": "media/bdv2fail/1c99f5cb2750d538b55f158b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0452/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 452, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the upper left of the table", "subtask": "move the blue cloth to the upper left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9cefbfdf0e942418f2953e94.png"}, {"type": "image", "uri": "media/bdv2fail/f51c5244623bca86fe29bd22.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0453/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/19", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 453, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "19", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "fold the cloth from top right top to center of the cloth", "subtask": "fold the cloth from top right top to center of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ce0c9bccd3f3409777460a4.png"}, {"type": "image", "uri": "media/bdv2fail/9ce0c9bccd3f3409777460a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0454/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_23_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 454, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_23_train", "episode_id": "13", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "moves spatula the spatula from right side to the middle of the stove.", "subtask": "moves spatula the spatula from right side to the middle of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5bc6786c9daed88027f82a1c.png"}, {"type": "image", "uri": "media/bdv2fail/b45f2201d5242af289fe6654.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0456/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 456, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the steel bowl on the left side above the right burner.", "subtask": "move the steel bowl on the left side above the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ddd85083e2e2dfeb9a04114e.png"}, {"type": "image", "uri": "media/bdv2fail/265a75f3a2f747e74f9bf53f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0458/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_41_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 458, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_41_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6a8fe056cc5a0cd4a151e15.png"}, {"type": "image", "uri": "media/bdv2fail/b6a8fe056cc5a0cd4a151e15.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0460/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/57", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 460, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "57", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "move the cheese to the right of the yellow toy", "subtask": "move the cheese to the right of the yellow toy", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8ecd81906d7736b2c835391.png"}, {"type": "image", "uri": "media/bdv2fail/9f5941add05dd616c3c28968.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0463/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_91_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 463, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_91_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the green toy from the top edge of the cloth to the left of the table.", "subtask": "move the green toy from the top edge of the cloth to the left of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/399c74ecbdf93fd99c244ccb.png"}, {"type": "image", "uri": "media/bdv2fail/e8f418c4d9bdf0fc7f0a5305.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0465/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/24", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 465, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "24", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "move the green object from the right side of the burner into the silver pot and then remove it", "subtask": "move the green object from the right side of the burner into the silver pot and then remove it", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5d10f6284b38b7238966ff9.png"}, {"type": "image", "uri": "media/bdv2fail/b7af6187125aa804960e2271.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0466/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 466, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "green cloth had been moved to the top right side", "subtask": "green cloth had been moved to the top right side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/93a080f9c2761a5fc61ba8ba.png"}, {"type": "image", "uri": "media/bdv2fail/e44eaa2d560096fe3a65fe21.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0469/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 469, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the grey toy spatula to the middle of the stove on top of the dial.", "subtask": "move the grey toy spatula to the middle of the stove on top of the dial.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a31fb74daa73a2fb8e174500.png"}, {"type": "image", "uri": "media/bdv2fail/87cba07974b47a294654043f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0470/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 470, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "take the lion toy and place it below the drawer", "subtask": "take the lion toy and place it below the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/acd0111bd5e8882b17aef9c0.png"}, {"type": "image", "uri": "media/bdv2fail/df99b5cd92ae3cabefe1b28a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0474/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/41", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 474, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "41", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the cloth to right from left", "subtask": "move the cloth to right from left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}, {"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0475/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 475, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the red bell pepper to the yellow knife.", "subtask": "move the red bell pepper to the yellow knife.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/412a2e87e5db6f7db95f7a13.png"}, {"type": "image", "uri": "media/bdv2fail/debb5383eabe0f80cd09fb46.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0479/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 479, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the red pepper to the front left corner of the stove.", "subtask": "move the red pepper to the front left corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7e39879c4ac02d36882eeb3d.png"}, {"type": "image", "uri": "media/bdv2fail/8b4d481b8551e15402d0fa7e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0480/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_02_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 480, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_02_train", "episode_id": "0", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the red spoon on the stove.", "subtask": "put the red spoon on the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9b90dfd3c0b69edbcb498f2.png"}, {"type": "image", "uri": "media/bdv2fail/d62a13dca5f3b8c3811ba8d8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0483/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 483, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "remove the silver pot from the upper right of the burner.", "subtask": "remove the silver pot from the upper right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fd1dbf5eb2689da071e02103.png"}, {"type": "image", "uri": "media/bdv2fail/eac5862f417c6f70b855240b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0487/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_44_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 487, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_44_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "take the gray rabbit toy and put it on the left burner.", "subtask": "take the gray rabbit toy and put it on the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a9d7aa88a571d3824cc66924.png"}, {"type": "image", "uri": "media/bdv2fail/e7bcb9dd50b777b9601f76b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0490/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 490, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the yellow dish drying rack to the sink from the plastic pink cup", "subtask": "put the yellow dish drying rack to the sink from the plastic pink cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c583f095c9071d4e3bca7aa.png"}, {"type": "image", "uri": "media/bdv2fail/ed8be19109dc9a2b044da7aa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0491/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 491, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "move the orange towel to the left of the stove.", "subtask": "move the orange towel to the left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/436dd5a3bb58b32a994ec871.png"}, {"type": "image", "uri": "media/bdv2fail/4f9f0a8faf67213d382c824e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0493/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_03_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 493, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_03_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the top of the stove", "subtask": "move the blue cloth to the top of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/817729855a101b7d59c24536.png"}, {"type": "image", "uri": "media/bdv2fail/e0b556772580d5bc06d71731.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0494/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 494, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "take the orange towel to the yellow plastic toy corn.", "subtask": "take the orange towel to the yellow plastic toy corn.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6740ab92c2bbc55198313fcd.png"}, {"type": "image", "uri": "media/bdv2fail/ab335c7f24fe59428a940c30.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0497/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 497, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the pot to the yellow spatula", "subtask": "move the pot to the yellow spatula", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}, {"type": "image", "uri": "media/bdv2fail/1cb02cfb13d4475d5012d2f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0506/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 506, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the cloth from left to right.", "subtask": "move the cloth from left to right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/97de6df2eb6720e14986adb1.png"}, {"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0507/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 507, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9fbbca24a8a9fbb667d1f19c.png"}, {"type": "image", "uri": "media/bdv2fail/5a3f2d55fd39673d4e40df96.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0509/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 509, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 215, "role": "error_localization"}}
{"state": {"task": "put the rice ball to the upper right side of the stove", "subtask": "put the rice ball to the upper right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f7e28b5ff3bdd992da8300dd.png"}, {"type": "image", "uri": "media/bdv2fail/a507cbaa3568c405e61f4525.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0510/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_13_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 510, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_13_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the blue fork in the bottom right of the burner", "subtask": "put the blue fork in the bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6e02e143d34bae3641ab9abe.png"}, {"type": "image", "uri": "media/bdv2fail/45bfa177faba36b7c7453f0e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0514/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_19_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 514, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_19_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the blue brush to the upper side of the toy sink next to the yellow cloth.", "subtask": "move the blue brush to the upper side of the toy sink next to the yellow cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0fbf5aaeb262929ab97488be.png"}, {"type": "image", "uri": "media/bdv2fail/4ee3570fad61e966def81307.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0515/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 515, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "fold the cloth from left to right", "subtask": "fold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d8b2e12e74e93a6cd74ca867.png"}, {"type": "image", "uri": "media/bdv2fail/d8b2e12e74e93a6cd74ca867.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0516/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 516, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train", "episode_id": "4", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "remove the corn from the upper right side of the burner", "subtask": "remove the corn from the upper right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e6fd821f7170f246601b6de3.png"}, {"type": "image", "uri": "media/bdv2fail/05f00798810589c29fe22dd1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0522/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_79_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 522, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_79_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the little box to the top left side of the stove", "subtask": "move the little box to the top left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1af61238df201834801aaf5f.png"}, {"type": "image", "uri": "media/bdv2fail/b7736924a83e75ea82de41f8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0524/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train/19", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 524, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train", "episode_id": "19", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the upper part of the stove", "subtask": "move the blue cloth to the upper part of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f25774d0c4ca9ff203f4981.png"}, {"type": "image", "uri": "media/bdv2fail/9f25774d0c4ca9ff203f4981.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0526/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 526, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "9", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the green object into the silver pot", "subtask": "move the green object into the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}, {"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0527/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 527, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "take the green sponge and move it on the upper left side of the stove.", "subtask": "take the green sponge and move it on the upper left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/67b970b3f09bc7683716b71b.png"}, {"type": "image", "uri": "media/bdv2fail/de88608fdd196c1e1e023139.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0529/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_75_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 529, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_75_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the blue towel to the upper right corner of the stove.", "subtask": "move the blue towel to the upper right corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b01bba73709b7ae77c93a420.png"}, {"type": "image", "uri": "media/bdv2fail/75a7d6d204701ba42c831d11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0530/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 530, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "close tthe drawer", "subtask": "close tthe drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc32637edbb787fb9de025ed.png"}, {"type": "image", "uri": "media/bdv2fail/bc32637edbb787fb9de025ed.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0532/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 532, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train", "episode_id": "11", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "move the red tomato to the back right corner of stove above cloth.", "subtask": "move the red tomato to the back right corner of stove above cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec08ffda626cede8d3d31f6e.png"}, {"type": "image", "uri": "media/bdv2fail/6927e1e015956cde7b4f8004.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0533/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_09_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 533, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_09_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the towel to the top side of the left burner.", "subtask": "move the towel to the top side of the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4e564177ef88ca6e8a84884.png"}, {"type": "image", "uri": "media/bdv2fail/e15609f091788e8448e94b2d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0535/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_02_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 535, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_02_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put spatula next to but not on plate sink", "subtask": "put spatula next to but not on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9f3ffe0b1b248aeee46759a.png"}, {"type": "image", "uri": "media/bdv2fail/204d280495a826912b8bed64.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0536/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 536, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "open the wooden box", "subtask": "open the wooden box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e7818742f29b833bbe529b26.png"}, {"type": "image", "uri": "media/bdv2fail/682fa35cf314c6a8ad3261b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0537/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_14_train/39", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 537, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_14_train", "episode_id": "39", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "remove the pot from the blue cloth.", "subtask": "remove the pot from the blue cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac72daba427d0cc5f6b9cdef.png"}, {"type": "image", "uri": "media/bdv2fail/ac72daba427d0cc5f6b9cdef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0538/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_68_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 538, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_68_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the toy microwave", "subtask": "move the toy microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/111fc4b791dad864853861ed.png"}, {"type": "image", "uri": "media/bdv2fail/90ddf693fad1497e361eb499.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0540/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 540, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "move the green cloth from left to right and place it on the black faucet", "subtask": "move the green cloth from left to right and place it on the black faucet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d5da668f57dbae50dbca495.png"}, {"type": "image", "uri": "media/bdv2fail/9fd3b47bcf9d897c052e765e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0541/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 541, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the red rectangular block below the ice cream toy", "subtask": "move the red rectangular block below the ice cream toy", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/38407db3eed87731a5ea2399.png"}, {"type": "image", "uri": "media/bdv2fail/b1a316f25cf30f886e3b4547.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0542/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/29", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 542, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "29", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the orange towel close by but not on the lower part of stove", "subtask": "move the orange towel close by but not on the lower part of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6f1bdfa08248426409a98767.png"}, {"type": "image", "uri": "media/bdv2fail/35511d4afd2354679282cf24.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0543/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 543, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "put the blue toy fork to the upper right burner from the upper left burner.", "subtask": "put the blue toy fork to the upper right burner from the upper left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e82556077fc148d1ff24724f.png"}, {"type": "image", "uri": "media/bdv2fail/57ed1961ac71af6086f633bd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0544/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 544, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from right to left", "subtask": "unfold the cloth from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e6128aff8323ad88f97404a.png"}, {"type": "image", "uri": "media/bdv2fail/3e6128aff8323ad88f97404a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0545/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 545, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "11", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "put the red tomatoes basket from counter to sink", "subtask": "put the red tomatoes basket from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/57578533979e054da972be29.png"}, {"type": "image", "uri": "media/bdv2fail/ed028fcec376804e6ce64226.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0546/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/24", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 546, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "24", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the green spatula to the right bottom side of the table.", "subtask": "move the green spatula to the right bottom side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a1f5ee1edc90360a984df94d.png"}, {"type": "image", "uri": "media/bdv2fail/1bfbfd8368f54b601bf91d36.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0550/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 550, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth upward the stove", "subtask": "move the yellow cloth upward the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52938137d441f4189764d1d7.png"}, {"type": "image", "uri": "media/bdv2fail/52938137d441f4189764d1d7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0554/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train/20", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 554, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train", "episode_id": "20", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the silver to the right of burner", "subtask": "move the silver to the right of burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e2247496372030225d89f7d5.png"}, {"type": "image", "uri": "media/bdv2fail/2f5f6b1c11dc209d11d071ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0555/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_39_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 555, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_39_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "put the pot on the bottom right of the stove", "subtask": "put the pot on the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d426044ecd9e4dfa01e60abf.png"}, {"type": "image", "uri": "media/bdv2fail/3e4f07afcf7b26a3ead73cdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0556/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 556, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the upper left side of the table", "subtask": "move the silver pot to the upper left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/99672f5be4f08bd9cc9845ed.png"}, {"type": "image", "uri": "media/bdv2fail/99672f5be4f08bd9cc9845ed.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0557/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 557, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "remove the lid from the silver bowl", "subtask": "remove the lid from the silver bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ee7766f25a37ddd4e16518e.png"}, {"type": "image", "uri": "media/bdv2fail/9ee7766f25a37ddd4e16518e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0560/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 560, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the green cloth to the bowl", "subtask": "move the green cloth to the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a375d4028dcc33f4bb9e969c.png"}, {"type": "image", "uri": "media/bdv2fail/bbadef55695967a32a6881f8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0561/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 561, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the blue towel to the top side of the faucet.", "subtask": "move the blue towel to the top side of the faucet.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9189bbaba8039f2d7f3760b9.png"}, {"type": "image", "uri": "media/bdv2fail/fa20c9e1cf504b31724106c2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0565/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 565, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put cup close to but not on plate", "subtask": "put cup close to but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d70567d9bf2c993a715fe709.png"}, {"type": "image", "uri": "media/bdv2fail/3677a790ecb1ad38d831c2e2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0566/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/24", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 566, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "24", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the spatula and place it just behind the orange peel.", "subtask": "move the spatula and place it just behind the orange peel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79e26cf73d134fe9b9181273.png"}, {"type": "image", "uri": "media/bdv2fail/d7260464af59a8e5e633d2cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0568/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 568, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the cloth with a red circle from right to left", "subtask": "move the cloth with a red circle from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/924374499e8937a81bdd5a4a.png"}, {"type": "image", "uri": "media/bdv2fail/74681f140d9c5ea2361a859f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0571/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train/19", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 571, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train", "episode_id": "19", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put the silver pot on the stove aer", "subtask": "put the silver pot on the stove aer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3234185a0bfa97411435a2a5.png"}, {"type": "image", "uri": "media/bdv2fail/000348d821b8723baadab927.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0573/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 573, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the green cloth to the bottom left corner of the stove", "subtask": "move the green cloth to the bottom left corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a38afdf6ee867f5d8a77f241.png"}, {"type": "image", "uri": "media/bdv2fail/dc3a4c42331e380a39ced400.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0576/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_110_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 576, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_110_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the yellow spatula to back right corner of stove top.", "subtask": "move the yellow spatula to back right corner of stove top.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/80ae2d3150b7a5723e0f8276.png"}, {"type": "image", "uri": "media/bdv2fail/522ed9bfbb3f41aaed804c19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0577/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_69_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 577, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_69_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "take the blue cloth to the left side of the burner from the right.", "subtask": "take the blue cloth to the left side of the burner from the right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1220b0cd25acbee2da4bebc8.png"}, {"type": "image", "uri": "media/bdv2fail/50b596b2e642d75615e7fccd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0581/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 581, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "pick up the silver pot and put it at the opposite corner of the stove", "subtask": "pick up the silver pot and put it at the opposite corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6c04c9a4660d662a4d09965b.png"}, {"type": "image", "uri": "media/bdv2fail/0cfe07136d1f07f9bc3de241.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0582/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 582, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "3", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "put the yellow doll on the red tomato", "subtask": "put the yellow doll on the red tomato", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/223ec694abb51b4ba5e33564.png"}, {"type": "image", "uri": "media/bdv2fail/64bd04c75ab8efcf838951dc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0584/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train/24", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 584, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train", "episode_id": "24", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the upper left edge of the stove.", "subtask": "move the silver pot to the upper left edge of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/37fbe2530f19896e16872ecd.png"}, {"type": "image", "uri": "media/bdv2fail/8599a6a4a54f0368f2d3c6e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0586/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 586, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from left to right", "subtask": "unfold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/74c518c19dc064e5d76a3fe4.png"}, {"type": "image", "uri": "media/bdv2fail/74c518c19dc064e5d76a3fe4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0592/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_12_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 592, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_12_train", "episode_id": "4", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the green fork to right of the silver pot", "subtask": "move the green fork to right of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4444af3ac5316c8102f15350.png"}, {"type": "image", "uri": "media/bdv2fail/ddd74a1844da9611a7fbf8b5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0593/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 593, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "remove a blue stuffed animal from the drawer.", "subtask": "remove a blue stuffed animal from the drawer.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/820d263ae01a56b5face038e.png"}, {"type": "image", "uri": "media/bdv2fail/fd83b81123d86595943ccc8d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0595/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 595, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "put the brown lid on the black pot", "subtask": "put the brown lid on the black pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e15ddef653eba328a364d00.png"}, {"type": "image", "uri": "media/bdv2fail/74934245d7b66974af3ff857.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0597/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 597, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "place the capsicum in the black pot", "subtask": "place the capsicum in the black pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/96837ab6c4383dc63e004864.png"}, {"type": "image", "uri": "media/bdv2fail/4c107d8927d96817d4e73b39.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0598/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_104_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 598, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_104_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the chocolate bar to the bottom of the left burner.", "subtask": "move the chocolate bar to the bottom of the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/275543f16e4856019bd6e0b5.png"}, {"type": "image", "uri": "media/bdv2fail/cee345aef808e5192b82a4f4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0603/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 603, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the carrot to the top side of the table.", "subtask": "move the carrot to the top side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dc23529253d7e8c92a258452.png"}, {"type": "image", "uri": "media/bdv2fail/0102cb17f61e200d318d78cc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0605/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_18_train/31", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 605, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_18_train", "episode_id": "31", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the steel pot close by but not on the left burner", "subtask": "put the steel pot close by but not on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4748fc852e7a5980f612b5f.png"}, {"type": "image", "uri": "media/bdv2fail/415ab6ceaf266d9719e4107b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0606/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_25_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 606, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_25_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the cloth to the center from top left of the white wall", "subtask": "move the cloth to the center from top left of the white wall", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c3b618e18d13e3431ee8bba4.png"}, {"type": "image", "uri": "media/bdv2fail/c880ccf918024d434d07cb4a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0607/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_02_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 607, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_02_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "put the corn toy yellow toy on the orange sponge", "subtask": "put the corn toy yellow toy on the orange sponge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2582749ce00c830a17349954.png"}, {"type": "image", "uri": "media/bdv2fail/8acbf18fd640c27d7a64daa7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0609/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 609, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "remove blueberries from plate sink", "subtask": "remove blueberries from plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cb2ebef068c027d40732b148.png"}, {"type": "image", "uri": "media/bdv2fail/cb2ebef068c027d40732b148.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0610/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 610, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "12", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "move the can at the left side of the burner", "subtask": "move the can at the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc01f813c13beb6cbcd794e6.png"}, {"type": "image", "uri": "media/bdv2fail/43a134743df93288cc3e3c78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0612/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 612, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the egg to the bottom right corner of the kitchen", "subtask": "move the egg to the bottom right corner of the kitchen", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/67acc55971616a0aef8883e4.png"}, {"type": "image", "uri": "media/bdv2fail/54171914038640c7a2485042.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0616/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_05_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 616, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_05_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "pick up the red spatula and place it above the right burner.", "subtask": "pick up the red spatula and place it above the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/16d516b28fb0006d8c2e519c.png"}, {"type": "image", "uri": "media/bdv2fail/f48a4c7c67434dc82eace01a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0618/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 618, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth from the middle front of the burners to behind the left side of the burners.", "subtask": "move the yellow cloth from the middle front of the burners to behind the left side of the burners.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b13a7904f6d38c43dac100f.png"}, {"type": "image", "uri": "media/bdv2fail/120004508eca18c9559e4284.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0621/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 621, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 251, "role": "error_localization"}}
{"state": {"task": "move the silver bowl to the lower side of the table.", "subtask": "move the silver bowl to the lower side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fc170c5286f9afa88eed2630.png"}, {"type": "image", "uri": "media/bdv2fail/74fd4002e211caa8d7b6ab98.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0622/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 622, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "cover the metal clamp with the pot.", "subtask": "cover the metal clamp with the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4fba0a9455049f1be4923ad8.png"}, {"type": "image", "uri": "media/bdv2fail/f2f20c1afa362c069a4dafdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0624/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 624, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "put silver pot at the bottom right side of the burner near the left switch", "subtask": "put silver pot at the bottom right side of the burner near the left switch", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eb090f9d4f08ae111b1598d7.png"}, {"type": "image", "uri": "media/bdv2fail/266a0024b2e61703d1eafbf9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0625/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 625, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the cloth from left to right", "subtask": "move the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/db5d032dab6631088ee96ee0.png"}, {"type": "image", "uri": "media/bdv2fail/94b20ac30d3e2f4726752c5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0626/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 626, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1298afc119b4d6883b0d5e00.png"}, {"type": "image", "uri": "media/bdv2fail/1298afc119b4d6883b0d5e00.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0628/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 628, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train", "episode_id": "0", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the blue spoon to the bottom right of the stove", "subtask": "move the blue spoon to the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/355752f3145a544ae2844058.png"}, {"type": "image", "uri": "media/bdv2fail/355752f3145a544ae2844058.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0633/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 633, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "1", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "push the blue cloth to the upper left of the burner", "subtask": "push the blue cloth to the upper left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3977181724dce9d999d17843.png"}, {"type": "image", "uri": "media/bdv2fail/5e63ddc3c4c04ec528c27256.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0634/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 634, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "pick up the toolbox.", "subtask": "pick up the toolbox.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/24557c4a225f8c8920d0ff1a.png"}, {"type": "image", "uri": "media/bdv2fail/df7c432e5f62f58e82974b38.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0635/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 635, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "put the orange cloth on the right stove", "subtask": "put the orange cloth on the right stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1952198e9c8e29d312d672f2.png"}, {"type": "image", "uri": "media/bdv2fail/6eba7c890cadea21880a9e77.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0636/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 636, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the red object from the top of the cloth to the table", "subtask": "move the red object from the top of the cloth to the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4099d5b9200d553d4785f374.png"}, {"type": "image", "uri": "media/bdv2fail/4099d5b9200d553d4785f374.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0637/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_07_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 637, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_07_train", "episode_id": "17", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the green spatula to the right side of the left burner", "subtask": "move the green spatula to the right side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d68a7d8391478253271339d3.png"}, {"type": "image", "uri": "media/bdv2fail/6300092461b3fe105761e809.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0643/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 643, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the yellow banana to the top right of the stove", "subtask": "move the yellow banana to the top right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d0a0d4915cc35d19b06387c5.png"}, {"type": "image", "uri": "media/bdv2fail/21d559a5f7975187f6fbe080.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0644/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 644, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the red bell pepper into bottom left of the burner", "subtask": "move the red bell pepper into bottom left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c829780dbb0b036c7ea0e29.png"}, {"type": "image", "uri": "media/bdv2fail/4270220db5bbf7711277da5e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0648/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_78_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 648, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_78_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the purple towel to the lower right side of the burner.", "subtask": "move the purple towel to the lower right side of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6f09b245e433adf0e664da97.png"}, {"type": "image", "uri": "media/bdv2fail/1e63651c2af4ee80b6b74a25.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0650/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 650, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the pot to the top left of the stove.", "subtask": "move the pot to the top left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec63d22963aa70f9b0318337.png"}, {"type": "image", "uri": "media/bdv2fail/2bb0eeccacc3bab266cab577.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0651/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 651, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the blue cup into the basket", "subtask": "put the blue cup into the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/618f735c173259d70592d0ef.png"}, {"type": "image", "uri": "media/bdv2fail/6863c54f5074103a8182f723.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0654/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/43", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 654, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "43", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a46a657d90cfb88bc5dbc5ae.png"}, {"type": "image", "uri": "media/bdv2fail/9d4f3f805e9e2e9d6e60adcc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0657/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 657, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 215, "role": "error_localization"}}
{"state": {"task": "move the metal bowl to the middle upper part of the table.", "subtask": "move the metal bowl to the middle upper part of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/29ffa4ac92ed3cee28963bcd.png"}, {"type": "image", "uri": "media/bdv2fail/59ccd025920406eddadf2f5b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0658/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 658, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the green towel from the bottom left to the top right of the burner.", "subtask": "move the green towel from the bottom left to the top right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6f5466de30f17f1ff35785f8.png"}, {"type": "image", "uri": "media/bdv2fail/7b36c36489b429f93624d2a6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0659/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_00_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 659, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_00_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "place the silver bowl on the stove", "subtask": "place the silver bowl on the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e48fc69974c490a71851ca31.png"}, {"type": "image", "uri": "media/bdv2fail/54646820b26a879c52cb0706.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0662/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_37_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 662, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_37_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl sink", "subtask": "put spoon in bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc645d1faf0738f842573afc.png"}, {"type": "image", "uri": "media/bdv2fail/bc645d1faf0738f842573afc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0664/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/24", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 664, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "24", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 219, "role": "error_localization"}}
{"state": {"task": "take the silver bowl and put the red object (tomato) on top of it.", "subtask": "take the silver bowl and put the red object (tomato) on top of it.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/187ca733a71978cd371599da.png"}, {"type": "image", "uri": "media/bdv2fail/51cb73c834f95be41d557a8c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0666/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_39_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 666, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_39_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "move the silver with blue handle spoon to the left side of the purple towel", "subtask": "move the silver with blue handle spoon to the left side of the purple towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/05b563a522e43973b4af866f.png"}, {"type": "image", "uri": "media/bdv2fail/cf4da7c1d8f61b80a21df127.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0668/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 668, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the grey sink to the middle of the burner from the right burner.", "subtask": "move the grey sink to the middle of the burner from the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7557e94fe58cabe8eb5e0db5.png"}, {"type": "image", "uri": "media/bdv2fail/064933e5ed3615b4131fd62f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0669/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 669, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "move the silver screw from left corner of the table to the top of the table.", "subtask": "move the silver screw from left corner of the table to the top of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/080af74866c974dbf4e2604f.png"}, {"type": "image", "uri": "media/bdv2fail/2a0e138e25f7a006347ac757.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0670/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 670, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "put cup in sink", "subtask": "put cup in sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/66610af9e667ac64d0de58e3.png"}, {"type": "image", "uri": "media/bdv2fail/33fe5830f6665ab65e9d49d6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0677/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/29", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 677, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "29", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "move the yellow spoon to the bottom left side of the stove.", "subtask": "move the yellow spoon to the bottom left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c8840a5746827bb7986e84b.png"}, {"type": "image", "uri": "media/bdv2fail/838eff72ab2968abc857c4cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0678/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 678, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the green object towel in the silver pot", "subtask": "put the green object towel in the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a579474f3d1b86375cbf1b68.png"}, {"type": "image", "uri": "media/bdv2fail/f5009d123f10282ba5b8ac73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0682/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_74_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 682, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_74_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "remove the corn from the silver pot", "subtask": "remove the corn from the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8ea5130969d36de31f2e4cd.png"}, {"type": "image", "uri": "media/bdv2fail/f8ea5130969d36de31f2e4cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0683/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 683, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "9", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "put the yellow towel near but not on the left burner", "subtask": "put the yellow towel near but not on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70c5fe5722bd30d78ae38041.png"}, {"type": "image", "uri": "media/bdv2fail/bc459149bd15a4f79827d310.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0684/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 684, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put banana in colander", "subtask": "put banana in colander", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4cce6af132a4d9ef7946e6e3.png"}, {"type": "image", "uri": "media/bdv2fail/4cce6af132a4d9ef7946e6e3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0686/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 686, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train", "episode_id": "4", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 219, "role": "error_localization"}}
{"state": {"task": "put eggplant beside but not on plate", "subtask": "put eggplant beside but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/07fac803b51973602f59c646.png"}, {"type": "image", "uri": "media/bdv2fail/8c14bccdab0c349ec99ce9b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0689/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 689, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the cloth to top from bottom", "subtask": "move the cloth to top from bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1af6cec5704137455abc579e.png"}, {"type": "image", "uri": "media/bdv2fail/c46bd3da739ec4bfb2c2403f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0690/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 690, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the yellow towel left to right.", "subtask": "move the yellow towel left to right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a98c8a90fe2b4bba010eddaa.png"}, {"type": "image", "uri": "media/bdv2fail/d2a90ee89fa450e34bc4ea7e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0692/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_94_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 692, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_94_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "pick the yellow and white knife and put it on the right side of the hob", "subtask": "pick the yellow and white knife and put it on the right side of the hob", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bbb6dcd64df2e14cadebc265.png"}, {"type": "image", "uri": "media/bdv2fail/7e14d152bc20fb7e8286c955.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0693/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 693, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "move orange cloth to the centre of the yellow toy carrot cone", "subtask": "move orange cloth to the centre of the yellow toy carrot cone", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8ca5d12b9780e495e5a7e395.png"}, {"type": "image", "uri": "media/bdv2fail/4415a60c56051ad8812520f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0694/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_11_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 694, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_11_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the carrot from the left to the right and put it in top of the green cloth.", "subtask": "move the carrot from the left to the right and put it in top of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/433a332fedef5d025d4d3bce.png"}, {"type": "image", "uri": "media/bdv2fail/bf69525ae0fc9e844c647e73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0701/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_28_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 701, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_28_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 245, "role": "error_localization"}}
{"state": {"task": "move the metal strainer to the lower right of the table.", "subtask": "move the metal strainer to the lower right of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86753477bfea123aa804406e.png"}, {"type": "image", "uri": "media/bdv2fail/ec0b1f8275b802d6fdd3d2b5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0702/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 702, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the yellow corn toy between the burner and the sink.", "subtask": "move the yellow corn toy between the burner and the sink.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b08094972edd79ade776c1e8.png"}, {"type": "image", "uri": "media/bdv2fail/18d057d5ab58fd5e66790e2e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0704/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_59_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 704, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_59_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the knob to the left burner from the right burner.", "subtask": "move the knob to the left burner from the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c0542e1be9affae188b8156.png"}, {"type": "image", "uri": "media/bdv2fail/83a77635946d79fb0807eca6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0706/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_04_train/20", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 706, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_04_train", "episode_id": "20", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the top of the pot.", "subtask": "move the blue cloth to the top of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d2787abaf5b3ee5d5292825.png"}, {"type": "image", "uri": "media/bdv2fail/4b0a158f06080203831bbff3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0707/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 707, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the red plastic fork on top of the green cloth.", "subtask": "move the red plastic fork on top of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7089ec6ffd01b8ca61834720.png"}, {"type": "image", "uri": "media/bdv2fail/f7b5747946910c69979765a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0708/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 708, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the knob-like object to the right side of the table", "subtask": "move the knob-like object to the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ecdfd201f556737580f33240.png"}, {"type": "image", "uri": "media/bdv2fail/091ce238a2d40c0066dd6f26.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0709/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 709, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the red toy scrubbing brush and place it on top of the orange cloth.", "subtask": "move the red toy scrubbing brush and place it on top of the orange cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7461e053a5b6e69086656776.png"}, {"type": "image", "uri": "media/bdv2fail/58274a51f9b000ffa42a7816.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0710/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 710, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "move the toy camera to the right side of the burner.", "subtask": "move the toy camera to the right side of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c42b354c174671d99573e52.png"}, {"type": "image", "uri": "media/bdv2fail/3092ccfbd3da8bc531aac88b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0712/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 712, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from top left to bottom right", "subtask": "unfold the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f9aace0806cc7c58e7928e38.png"}, {"type": "image", "uri": "media/bdv2fail/f9aace0806cc7c58e7928e38.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0718/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_18_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 718, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_18_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the blue plastic toy from left to right", "subtask": "move the blue plastic toy from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1f434e508d902c274259396c.png"}, {"type": "image", "uri": "media/bdv2fail/4569e7869ae8d0624a3d1055.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0719/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 719, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the blue blob on the bottom to the opposite edge of the right burner.", "subtask": "move the blue blob on the bottom to the opposite edge of the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/61e0c06e9c0217d1efecf27f.png"}, {"type": "image", "uri": "media/bdv2fail/d36860e69368c7b98a43e5e1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0721/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_77_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 721, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_77_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the top left side of the table.", "subtask": "move the silver pot to the top left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ae970fe18df2b8bd6a9de3c4.png"}, {"type": "image", "uri": "media/bdv2fail/3216c5703c93ea25b751d76c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0722/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 722, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the yellow spatula to the left top corner of the burner.", "subtask": "move the yellow spatula to the left top corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5921aeefd3e35236b3f68ab.png"}, {"type": "image", "uri": "media/bdv2fail/df7185d51294831a7bb0f6f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0726/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_55_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 726, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_55_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the brown and white stuffed animal to the left side of the table.", "subtask": "move the brown and white stuffed animal to the left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ad068cff729074332a509944.png"}, {"type": "image", "uri": "media/bdv2fail/16043f1b7d3b61bfab9dd8c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0728/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 728, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "move the red pepper pot on the bottom right corner of the burner.", "subtask": "move the red pepper pot on the bottom right corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b12600085f1b3fedb1f06c91.png"}, {"type": "image", "uri": "media/bdv2fail/c6b02a855e0038a796ef6352.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0729/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_78_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 729, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_78_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "put the yellow and orange spoon on the silver pot", "subtask": "put the yellow and orange spoon on the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0d01670dcbeb903c6f15302d.png"}, {"type": "image", "uri": "media/bdv2fail/00a7285242f1286e37a75331.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0731/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 731, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "take the pot and put it around but not on the green towel", "subtask": "take the pot and put it around but not on the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d3163e550be9737fab30421.png"}, {"type": "image", "uri": "media/bdv2fail/ae181b53b43ecae1217cc5e0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0732/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 732, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the pot to the right side of the table", "subtask": "move the pot to the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/609241498e32b42864514228.png"}, {"type": "image", "uri": "media/bdv2fail/760abfdd687db263eca229a3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0733/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 733, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the half orange in the half orange", "subtask": "put the half orange in the half orange", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/113282c5a87d7796e1031a58.png"}, {"type": "image", "uri": "media/bdv2fail/e6c67c205c9f478ef5da1c0c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0734/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 734, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "remove the banana from the napkin from the right side of the table", "subtask": "remove the banana from the napkin from the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f6334491f709ce9f247bad63.png"}, {"type": "image", "uri": "media/bdv2fail/91d129be15c2df2fd7b2cbe9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0737/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 737, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "place the blue spoon on the right side of the pot.", "subtask": "place the blue spoon on the right side of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/25f7eb5ce2919263dd6d6726.png"}, {"type": "image", "uri": "media/bdv2fail/a354e99bda4ac74498ef0d6a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0741/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 741, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the blue handle below the right burner", "subtask": "move the blue handle below the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/314bdf507e2340356d9737f0.png"}, {"type": "image", "uri": "media/bdv2fail/87f3040a964040a0fd38dc9b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0745/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_26_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 745, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_26_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the cloth from the middle to the top edge of the table.", "subtask": "move the cloth from the middle to the top edge of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f98fa713ba1b7dfdeb87df89.png"}, {"type": "image", "uri": "media/bdv2fail/c52c5f804f16982ef87d98b9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0746/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 746, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the blue silver spoon from right side of the burner to the center of the induction stove", "subtask": "move the blue silver spoon from right side of the burner to the center of the induction stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3cd21797217244e8c8f8aa71.png"}, {"type": "image", "uri": "media/bdv2fail/bd4a34150bbda26b93e0d803.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0747/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 747, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 245, "role": "error_localization"}}
{"state": {"task": "move the silver pot on the left side of the burner", "subtask": "move the silver pot on the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5e934bf3cd6b4baa6be40f1.png"}, {"type": "image", "uri": "media/bdv2fail/bfa4813fda5f9fb4edcf1372.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0748/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 748, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the cloth to the left side of the table", "subtask": "move the cloth to the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3cd12630ab5252b9769291f4.png"}, {"type": "image", "uri": "media/bdv2fail/218df4d9b5ed37da1cfbe488.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0749/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_01_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 749, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_01_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the blue fork to the upper left side of the stove", "subtask": "move the blue fork to the upper left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f63b382af30b50c6c65c850.png"}, {"type": "image", "uri": "media/bdv2fail/4f63b382af30b50c6c65c850.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0752/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_10_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 752, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_10_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the black burner to the top left", "subtask": "move the black burner to the top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d62a62d280800b3bdf618ebb.png"}, {"type": "image", "uri": "media/bdv2fail/36ce2c30811922a2172f1f7c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0753/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_04_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 753, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_04_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "pick up blue pen and put close by but not into drawer", "subtask": "pick up blue pen and put close by but not into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cfa56063aa926abebc96b070.png"}, {"type": "image", "uri": "media/bdv2fail/5971535d2a6cfb8965d04602.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0754/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 754, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the metal bowl to the upper right corner of the cooker.", "subtask": "move the metal bowl to the upper right corner of the cooker.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/45a2bb8bd12c8c54b1f02822.png"}, {"type": "image", "uri": "media/bdv2fail/9d4d5ffa08d28f97c0d6f272.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0757/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 757, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the green cloth to the right burner from the bottom left side of the stove.", "subtask": "move the green cloth to the right burner from the bottom left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0daa54e579b1052200f6164a.png"}, {"type": "image", "uri": "media/bdv2fail/bc4e76630abe1a643f2a2624.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0761/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_80_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 761, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_80_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "move the toy to the right side of the table.", "subtask": "move the toy to the right side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7a02bd12d15640b73901c947.png"}, {"type": "image", "uri": "media/bdv2fail/1c6648f50a3df4ce9c35e8c4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0762/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 762, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the toy mushroom to the right in front of the stove", "subtask": "move the toy mushroom to the right in front of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c83c88f550b82b62d6a1182f.png"}, {"type": "image", "uri": "media/bdv2fail/aa7c12fd9df6e4b0581bc6ea.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0765/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_32_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 765, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_32_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "remove cucumber from cup", "subtask": "remove cucumber from cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b3ca830dafcdf61fedee5ae.png"}, {"type": "image", "uri": "media/bdv2fail/7b3ca830dafcdf61fedee5ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0766/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/31", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 766, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "31", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 217, "role": "error_localization"}}
{"state": {"task": "move blue toy spoon over orange cloth on the right stove.", "subtask": "move blue toy spoon over orange cloth on the right stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/28ccd04f674bb1bce1a2ad4b.png"}, {"type": "image", "uri": "media/bdv2fail/471f97d879ed91e552cddd0d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0768/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 768, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "take the white spoon from the silver pot and put it on the lower left corner of the hob.", "subtask": "take the white spoon from the silver pot and put it on the lower left corner of the hob.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d37ddaed5284e4f411886492.png"}, {"type": "image", "uri": "media/bdv2fail/5b52a77d151af3108c19293d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0769/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_22_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 769, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_22_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "move the blue plastic toy to left from right", "subtask": "move the blue plastic toy to left from right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4569e7869ae8d0624a3d1055.png"}, {"type": "image", "uri": "media/bdv2fail/1f434e508d902c274259396c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0771/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 771, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the blue plastic spoon to the bottom of the left burner", "subtask": "move the blue plastic spoon to the bottom of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1383ffb9ad397d8e247dbc42.png"}, {"type": "image", "uri": "media/bdv2fail/812c35b0fe4614f3cbcab8ea.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0776/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_02_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 776, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_02_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the purple cloth next to the bowl next to the sushi.", "subtask": "move the purple cloth next to the bowl next to the sushi.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6dab21dc75a998985940193.png"}, {"type": "image", "uri": "media/bdv2fail/83f13841d65e1577a3943e40.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0777/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_17_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 777, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_17_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the silver pot on the lower left corner of the burner", "subtask": "move the silver pot on the lower left corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3f6f9f39bb25ea56d7c2dd7d.png"}, {"type": "image", "uri": "media/bdv2fail/87fe80173b1d745435613f00.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0778/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 778, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from middle to right", "subtask": "unfold the cloth from middle to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/647a91c634ef5e7e06815d78.png"}, {"type": "image", "uri": "media/bdv2fail/647a91c634ef5e7e06815d78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0779/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_12_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 779, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_12_train", "episode_id": "5", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the top of the blue curtains", "subtask": "move the silver pot to the top of the blue curtains", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91019a6c95a080684ca23b8b.png"}, {"type": "image", "uri": "media/bdv2fail/ab07e2b7eb87cd9daef52745.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0789/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 789, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "pick up scissors and put close to but not inside drawer", "subtask": "pick up scissors and put close to but not inside drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e965f9d8e06b966ff135b275.png"}, {"type": "image", "uri": "media/bdv2fail/2ffe84b552b721d6587847fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0791/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train/38", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 791, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train", "episode_id": "38", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the silver pot lid to the pot from the yellow cloth on the top right of the table", "subtask": "move the silver pot lid to the pot from the yellow cloth on the top right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9b31e86b8bb31773b6d23303.png"}, {"type": "image", "uri": "media/bdv2fail/8ccfe7756ea7cb41987b5ec2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0794/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/0", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 794, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "take the orange towel and move it to the left side of the burner.", "subtask": "take the orange towel and move it to the left side of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d6b6fc758392eb63c0c3f2fc.png"}, {"type": "image", "uri": "media/bdv2fail/8d48d02d3e4aacf580c2f92d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0795/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 795, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "picking the blue plastic item from the stove and place it to the left burner on the left side.", "subtask": "picking the blue plastic item from the stove and place it to the left burner on the left side.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a3c3168a4bc7d69cb2264d99.png"}, {"type": "image", "uri": "media/bdv2fail/f287190cff6cff02b0cbedde.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0796/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 796, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 249, "role": "error_localization"}}
{"state": {"task": "move cloth to the bottom right", "subtask": "move cloth to the bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/83e0ec47ae798338e36124f9.png"}, {"type": "image", "uri": "media/bdv2fail/7714c444e6c35953a15fd3d5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0797/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_08_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 797, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_08_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "put the hot dog on the right side of the blue spoon", "subtask": "put the hot dog on the right side of the blue spoon", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/36d55bca860aecd69a086080.png"}, {"type": "image", "uri": "media/bdv2fail/5fcdc4abff62660b2c2460e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0801/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 801, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the gray pot on the lower left side.", "subtask": "move the gray pot on the lower left side.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d5a5c43f38c89a29e6a15a69.png"}, {"type": "image", "uri": "media/bdv2fail/ca53019445c5da9fb337f630.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0804/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_88_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 804, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_88_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "take the white spoon and put it between the center of the burner", "subtask": "take the white spoon and put it between the center of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bcdf71b4c26de9e7e26f0c65.png"}, {"type": "image", "uri": "media/bdv2fail/8552aed0e825f00fd9a77869.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0805/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 805, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "moves the green star-shaped object to top.", "subtask": "moves the green star-shaped object to top.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fb1f7a299fc1a5a0c850d91f.png"}, {"type": "image", "uri": "media/bdv2fail/6e09576c6a4b866c8aed8fda.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0808/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 808, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "fold cloth from top left to bottom right", "subtask": "fold cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7895da3647a6d617a0a34853.png"}, {"type": "image", "uri": "media/bdv2fail/7895da3647a6d617a0a34853.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0810/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train/33", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 810, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train", "episode_id": "33", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "put the green toy underneath of purple cloth", "subtask": "put the green toy underneath of purple cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4fabef0ffb84d546246736c7.png"}, {"type": "image", "uri": "media/bdv2fail/1f2210baa147340cd6314b54.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0812/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 812, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the yellow spatula to top right of burner", "subtask": "move the yellow spatula to top right of burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c67cca250de9fd7b64d2f5dd.png"}, {"type": "image", "uri": "media/bdv2fail/a307de57467c1d3ddf3a6255.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0813/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_08_train/3", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 813, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_08_train", "episode_id": "3", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the can of fake food to the top right of the stove.", "subtask": "move the can of fake food to the top right of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e63a1c16f1c917b268df1620.png"}, {"type": "image", "uri": "media/bdv2fail/923ff77afbfae10648ac9d69.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0814/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 814, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "fold the cloth from the left to the right", "subtask": "fold the cloth from the left to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/551eb149f3976e4ed2af0701.png"}, {"type": "image", "uri": "media/bdv2fail/551eb149f3976e4ed2af0701.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0817/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_18_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 817, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_18_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "put the eggplant broccoli in the drawer", "subtask": "put the eggplant broccoli in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b7c699726c0764aa80b18728.png"}, {"type": "image", "uri": "media/bdv2fail/baaae7251684e6571a4a277c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0818/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_05_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 818, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_05_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "touch the banana in the drawer", "subtask": "touch the banana in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cf252df7b29567eafc88f8af.png"}, {"type": "image", "uri": "media/bdv2fail/c7311c62928a6c661f31a8b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0821/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 821, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the bottom right of the burner.", "subtask": "move the blue cloth to the bottom right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/417be5b92ba7ce76de67bba6.png"}, {"type": "image", "uri": "media/bdv2fail/4e6faddc9d115edea4ae61fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0823/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_103_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 823, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_103_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "moving the toy ball from the center bottom of two burner to center top.", "subtask": "moving the toy ball from the center bottom of two burner to center top.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/78912662b490240b07844e31.png"}, {"type": "image", "uri": "media/bdv2fail/ce5918e237f8e51bae7fbb1f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0824/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 824, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "take the yellow banana and put beside but not in the left bottom of the table", "subtask": "take the yellow banana and put beside but not in the left bottom of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a61f3f332984fa96881c32be.png"}, {"type": "image", "uri": "media/bdv2fail/fc9002a13b7ca9b2e6f7a693.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0827/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/26", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 827, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "26", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "move the yellow spoon to the bottom of the right burner and blue cloth.", "subtask": "move the yellow spoon to the bottom of the right burner and blue cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6209230ade8e2a33b4370109.png"}, {"type": "image", "uri": "media/bdv2fail/36ced3fe78a1599472814256.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0831/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_69_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 831, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_69_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "move the white and orange object around but not on the right burner", "subtask": "move the white and orange object around but not on the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d2041f4c15a4c9033cbdba7d.png"}, {"type": "image", "uri": "media/bdv2fail/fa47e5cd854727c48503f5fd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0834/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 834, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move silver pot to back left corner of stove top", "subtask": "move silver pot to back left corner of stove top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc5c86687fd9524e7c522f98.png"}, {"type": "image", "uri": "media/bdv2fail/e68ad6b23e94ea0074e1d640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0836/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_103_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 836, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_103_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the red spoon in the right burner", "subtask": "put the red spoon in the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a31a96a803cb2c74b124773.png"}, {"type": "image", "uri": "media/bdv2fail/3a31a96a803cb2c74b124773.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0837/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 837, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the top right side of the table", "subtask": "move the silver pot to the top right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ac07bb9fabdcee6bc274b0f.png"}, {"type": "image", "uri": "media/bdv2fail/3cf39080b46015547b130b17.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0838/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 838, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put the yellow knife on the orange cloth.", "subtask": "put the yellow knife on the orange cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0d5e34211222e84861c37f9d.png"}, {"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0842/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 842, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from right to left side", "subtask": "unfold the cloth from right to left side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d207839ff359e7890cf0c7da.png"}, {"type": "image", "uri": "media/bdv2fail/d207839ff359e7890cf0c7da.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0843/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 843, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "7", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth in the right side of the silver pot", "subtask": "move the yellow cloth in the right side of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b91bfd5c180545c3532b835f.png"}, {"type": "image", "uri": "media/bdv2fail/03362fe902c4e5de5b1c60d2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0845/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 845, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "put cup around but not on plate", "subtask": "put cup around but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9eba0e1cf2b8fca02b17a05a.png"}, {"type": "image", "uri": "media/bdv2fail/05e174d4bfb7f2c026e589f6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0846/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/30", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 846, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "30", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the green towel to upper left of the sink.", "subtask": "move the green towel to upper left of the sink.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/36945f57034a1427eb12cf0c.png"}, {"type": "image", "uri": "media/bdv2fail/8556f54b205885711dd44d21.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0848/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 848, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the yellow knife to the top of the can.", "subtask": "move the yellow knife to the top of the can.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ff76cb4da0d554edd36bd7c8.png"}, {"type": "image", "uri": "media/bdv2fail/bf53c4f8f687a9663acab521.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0849/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train/20", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 849, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train", "episode_id": "20", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the blue stuffed toy above the drawer", "subtask": "put the blue stuffed toy above the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8b9e16d4306697fcc8b37abf.png"}, {"type": "image", "uri": "media/bdv2fail/d3ee15198715f4a4f7465693.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0850/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/31", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 850, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "31", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "take the green broccoli off the silver pot", "subtask": "take the green broccoli off the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aacd5e9e67e82854bf9fcc71.png"}, {"type": "image", "uri": "media/bdv2fail/c3e1a84c5b54d1062518eb5d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0851/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 851, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move silver pot onto the left burner.", "subtask": "move silver pot onto the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2fc973adb787cd5ebf1b48dd.png"}, {"type": "image", "uri": "media/bdv2fail/14370880da5faf7909bfac78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0854/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 854, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the blue object to the upper left side of the right burner", "subtask": "move the blue object to the upper left side of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/946b113097009abfd7b9a8b2.png"}, {"type": "image", "uri": "media/bdv2fail/0bcaf4d1b95d74c883342cb3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0855/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_70_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 855, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_70_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the left top burner.", "subtask": "move the silver pot to the left top burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/51aa2ba54aa20671d6b510fe.png"}, {"type": "image", "uri": "media/bdv2fail/11dcf36cec255fe24d4be854.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0857/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_00_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 857, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_00_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "place the green object inside the red toy box", "subtask": "place the green object inside the red toy box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e91d0dd17e8e5cae5243902a.png"}, {"type": "image", "uri": "media/bdv2fail/6751c9bd7fa2124cc4ebd2fc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0861/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_01_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 861, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_01_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "put the red object on the left side of the table.", "subtask": "put the red object on the left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/68cd038d94edc6d07fb40bd4.png"}, {"type": "image", "uri": "media/bdv2fail/109070970ba60ad8fae0729f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0863/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 863, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "take the red knob to the bottom right of the burner.", "subtask": "take the red knob to the bottom right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/53789d6bb02a991fdcdeaad5.png"}, {"type": "image", "uri": "media/bdv2fail/8560fe0afeb1efd46483c92d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0864/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 864, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "pick up screwdriver and put into drawer", "subtask": "pick up screwdriver and put into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6a8614ceb4fc31e2b2a784ef.png"}, {"type": "image", "uri": "media/bdv2fail/2474d52ded7ccd4d94c80342.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0866/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_glue_and_put_into_drawer_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 866, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_glue_and_put_into_drawer_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc5ec89012aeb7227cfd52ae.png"}, {"type": "image", "uri": "media/bdv2fail/cc5ec89012aeb7227cfd52ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0871/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 871, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "put corn next to but not in bowl sink", "subtask": "put corn next to but not in bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/63ae23d93612a7ee2a93e2d3.png"}, {"type": "image", "uri": "media/bdv2fail/bdf054964393383338b02ae6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0872/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/43", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 872, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "43", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "moves the ribbon from right of the cloth to top", "subtask": "moves the ribbon from right of the cloth to top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/43eea7b49ab6e0a76025341e.png"}, {"type": "image", "uri": "media/bdv2fail/4ebbc64e9b733f0257e30a7e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0873/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/34", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 873, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "34", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put cup to counter from sink", "subtask": "put cup to counter from sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e871c8d4ae3e86b344448ec.png"}, {"type": "image", "uri": "media/bdv2fail/8e871c8d4ae3e86b344448ec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0877/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/26", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 877, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "26", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "remove the blue spoon from the right side of the burner", "subtask": "remove the blue spoon from the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3892f532518a48b8a2a86cdc.png"}, {"type": "image", "uri": "media/bdv2fail/3892f532518a48b8a2a86cdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0880/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/9", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 880, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "9", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the stove to the right side of the yellow cloth", "subtask": "move the stove to the right side of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ce2ab4922adfc005e76b76a3.png"}, {"type": "image", "uri": "media/bdv2fail/89c7a1c81e063b4eef1f0774.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0881/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_03_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 881, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_03_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the metal bowl with the robot and place it on the far edge of the stove.", "subtask": "move the metal bowl with the robot and place it on the far edge of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13beef7979ad5e47dac9fd39.png"}, {"type": "image", "uri": "media/bdv2fail/f30ec8d748400e79b73fc16a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0882/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_91_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 882, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_91_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "move a blue cloth to the bottom right side of the toy oven.", "subtask": "move a blue cloth to the bottom right side of the toy oven.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a73fbf24b4cd93e1bf761400.png"}, {"type": "image", "uri": "media/bdv2fail/dbafda0d494d99390f83bc12.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0883/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_03_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 883, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_03_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "take out the chocolate above the drawer and put it on the bottom center of the table", "subtask": "take out the chocolate above the drawer and put it on the bottom center of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c91ad779f194f082b891595c.png"}, {"type": "image", "uri": "media/bdv2fail/f9d59305309decdffbf7ebfb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0884/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_12_train/31", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 884, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_12_train", "episode_id": "31", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "move the silver pot on the lower right side", "subtask": "move the silver pot on the lower right side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ff7e7cc22766fcc3ff5fabd.png"}, {"type": "image", "uri": "media/bdv2fail/0463f22e3e28f84cc5da8021.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0886/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_87_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 886, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_87_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "grab the yellow square to the middle and moved it from the bottom left edge", "subtask": "grab the yellow square to the middle and moved it from the bottom left edge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/66b4d43bcb2a9242a54413f4.png"}, {"type": "image", "uri": "media/bdv2fail/44ce6fdadd8eeb0c8a1afdf9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0889/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 889, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "put blueberries close to but not on plate sink", "subtask": "put blueberries close to but not on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e927d5f7aba53a029baead04.png"}, {"type": "image", "uri": "media/bdv2fail/d64f39ba48d655ea0af3e8df.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0890/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 890, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "move the duck from right to left", "subtask": "move the duck from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f3e3872f7b27130175efb216.png"}, {"type": "image", "uri": "media/bdv2fail/f3e3872f7b27130175efb216.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0891/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/21", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 891, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "21", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the cloth from the top left of the table to the bottom right.", "subtask": "move the cloth from the top left of the table to the bottom right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3c00ab61e417dfcc122bd2fc.png"}, {"type": "image", "uri": "media/bdv2fail/f92c2170a61483ff8b412b20.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0892/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/17", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 892, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "put the silver pot on top of the silver spoon.", "subtask": "put the silver pot on top of the silver spoon.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0592cb5c5172e9ecc9d45806.png"}, {"type": "image", "uri": "media/bdv2fail/2e631394f8d985105fa1f975.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0895/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 895, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put cup near but not on plate", "subtask": "put cup near but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf27f77209ad7ef28e7779b0.png"}, {"type": "image", "uri": "media/bdv2fail/cf119044eef1f2d1c1f0beaf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0897/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/39", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 897, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "39", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "close microwave", "subtask": "close microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd143b1aff093837e28a456b.png"}, {"type": "image", "uri": "media/bdv2fail/3b1013bd49180562d6f01627.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0900/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_open_microwave_train/44", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 900, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_open_microwave_train", "episode_id": "44", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 213, "role": "error_localization"}}
{"state": {"task": "put cucumber close by but not in cup", "subtask": "put cucumber close by but not in cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dbce06c934d3d93ffac65ae1.png"}, {"type": "image", "uri": "media/bdv2fail/23b9766f583bf6909b474dc1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0902/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/34", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 902, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "34", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "place the purple figure at the top left of the wooden box", "subtask": "place the purple figure at the top left of the wooden box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/07af6127d9cb20e9964817d8.png"}, {"type": "image", "uri": "media/bdv2fail/dde9e92b963f81b80104fd8f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0903/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_12_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 903, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_12_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "remove the orange fish from the silver bowl", "subtask": "remove the orange fish from the silver bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91da124d73ab01a9507065c6.png"}, {"type": "image", "uri": "media/bdv2fail/91da124d73ab01a9507065c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0904/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 904, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the can near the small orange bowl.", "subtask": "move the can near the small orange bowl.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6d0dda7dc49da7a5ba4f0398.png"}, {"type": "image", "uri": "media/bdv2fail/bd48e75c8316045c8bf86c53.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0905/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_16_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 905, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_16_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "take the yellow banana out of the pot and put it underneath of the table", "subtask": "take the yellow banana out of the pot and put it underneath of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fdc49c975ddf45ddac198224.png"}, {"type": "image", "uri": "media/bdv2fail/7a22268f81ad109f10c1cdc9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0914/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 914, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "error_localization"}}
{"state": {"task": "take the yellow brush to the lower left side of the table from the upper center.", "subtask": "take the yellow brush to the lower left side of the table from the upper center.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e347eb1a228e05441507ea5c.png"}, {"type": "image", "uri": "media/bdv2fail/5637b2725d4a5353bc7d991c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0917/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 917, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "fold the cloth from top left to bottom right", "subtask": "fold the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50392ae8f1056625bc7c6066.png"}, {"type": "image", "uri": "media/bdv2fail/50392ae8f1056625bc7c6066.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0918/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_23_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 918, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_23_train", "episode_id": "8", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "put the blue cup into the basket", "subtask": "put the blue cup into the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6863c54f5074103a8182f723.png"}, {"type": "image", "uri": "media/bdv2fail/618f735c173259d70592d0ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0920/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/43", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 920, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "43", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth to the upper right of the stove", "subtask": "move the yellow cloth to the upper right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/87883a8c33ae5d84b5115a16.png"}, {"type": "image", "uri": "media/bdv2fail/1cfcc3d8a351f7dffa3a4055.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0921/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 921, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "place the blue spoon on the right side of the stove", "subtask": "place the blue spoon on the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/25f7eb5ce2919263dd6d6726.png"}, {"type": "image", "uri": "media/bdv2fail/a354e99bda4ac74498ef0d6a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0923/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 923, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "remove the green pepper from the towel", "subtask": "remove the green pepper from the towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f04cd54c46e4847cc8be5341.png"}, {"type": "image", "uri": "media/bdv2fail/791cd75e4d4e4b1edf8fcfe2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0924/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_05_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 924, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_05_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the stove to the bottom right of the silver pot", "subtask": "move the stove to the bottom right of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/686d4633d579b17543c66d56.png"}, {"type": "image", "uri": "media/bdv2fail/d93db97b19b0a4b0bb68ce90.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0928/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/15", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 928, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "remove the blue figure from the top edge of the cloth", "subtask": "remove the blue figure from the top edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dc4228d0728db133a5314b33.png"}, {"type": "image", "uri": "media/bdv2fail/dc4228d0728db133a5314b33.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0932/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/31", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 932, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "31", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "remove corn from bowl sink", "subtask": "remove corn from bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3124b5ddd6d73d5c63413db7.png"}, {"type": "image", "uri": "media/bdv2fail/3124b5ddd6d73d5c63413db7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0934/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/1", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 934, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "1", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 219, "role": "error_localization"}}
{"state": {"task": "move the yellow plastic spoon to the left of the metal clamp", "subtask": "move the yellow plastic spoon to the left of the metal clamp", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/76567c96ef3ae8612f0c2ee5.png"}, {"type": "image", "uri": "media/bdv2fail/3de0f83464560698d22cf8a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0937/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 937, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "take the green towel and move it to the top of the desk next to the cans", "subtask": "take the green towel and move it to the top of the desk next to the cans", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9cba7b20bc5a2d9962a08ae1.png"}, {"type": "image", "uri": "media/bdv2fail/c88451ad8f5471f714fe164b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0939/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train/16", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 939, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "error_localization"}}
{"state": {"task": "move the red bowl to the top left corner of the stove.", "subtask": "move the red bowl to the top left corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b929de4c754f6db96d457213.png"}, {"type": "image", "uri": "media/bdv2fail/26e1990864c0db4bab305fc2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0940/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_40_train/8", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 940, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_40_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the toy blue fork to the left side of the stove", "subtask": "move the toy blue fork to the left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1987a877a7ba672d99f9285.png"}, {"type": "image", "uri": "media/bdv2fail/ec94647865e8d103ccc9b4b0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0941/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_32_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 941, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_32_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the blue plastic fork to the toy sink", "subtask": "move the blue plastic fork to the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aa67636d2834cd9744c6e590.png"}, {"type": "image", "uri": "media/bdv2fail/a9da7a85afe8afa40602f3b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0945/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 945, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the pot to the left burner.", "subtask": "move the pot to the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8c322d7c759d78937a33a526.png"}, {"type": "image", "uri": "media/bdv2fail/a2d6f6691330bcd5352cd0b9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0946/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 946, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move blue fork to the top right of burners", "subtask": "move blue fork to the top right of burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91d8e8347b3e3867bbd7d407.png"}, {"type": "image", "uri": "media/bdv2fail/91d8e8347b3e3867bbd7d407.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0950/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 950, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "error_localization"}}
{"state": {"task": "put the yellow knife on the grey knob", "subtask": "put the yellow knife on the grey knob", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ebd528d3388bf89c6be5de3e.png"}, {"type": "image", "uri": "media/bdv2fail/97e2d93377a76d85755cd7bb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0954/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_08_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 954, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_08_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the red plastic fork to the lower left side burner", "subtask": "move the red plastic fork to the lower left side burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4fbecb6bc530c1f36c14b5dc.png"}, {"type": "image", "uri": "media/bdv2fail/3779fdbb0ddbb8ec5e9cb1d1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0956/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_43_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 956, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_43_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the red fork to the right side of the stove", "subtask": "move the red fork to the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}, {"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0957/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 957, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "put the toy next to the red handle", "subtask": "put the toy next to the red handle", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a36e3a8487260877ca19c80.png"}, {"type": "image", "uri": "media/bdv2fail/8986315a67669650333507af.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0958/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train/36", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 958, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train", "episode_id": "36", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the yellow sponge to the back left from the front right corner.", "subtask": "move the yellow sponge to the back left from the front right corner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec02bd65b6583fd4c5d4b94a.png"}, {"type": "image", "uri": "media/bdv2fail/bffc7ba1b7cc212d6ed0a846.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0959/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/11", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 959, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from bottom left to top right", "subtask": "unfold the cloth from bottom left to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8c82a6dae95e1146a05c683c.png"}, {"type": "image", "uri": "media/bdv2fail/8c82a6dae95e1146a05c683c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0966/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_19_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 966, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_19_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "move the croissant and put it just below the croissant", "subtask": "move the croissant and put it just below the croissant", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/46dfac3b9883fc7eefdcd120.png"}, {"type": "image", "uri": "media/bdv2fail/9b18e3b01c23b57aa149f7a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0967/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_103_train/4", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 967, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_103_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "error_localization"}}
{"state": {"task": "move the toy pan to the upper side of the table", "subtask": "move the toy pan to the upper side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b10ba904844b8db885cc4738.png"}, {"type": "image", "uri": "media/bdv2fail/2fd61c74804401681bb1ced5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0968/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train/2", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 968, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the can around but not inside the lower side of the right burner", "subtask": "move the can around but not inside the lower side of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0c8de888bf45a9546a6c42bb.png"}, {"type": "image", "uri": "media/bdv2fail/f71f45f46f18b6ded34e72f4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0970/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 970, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom left of the burner", "subtask": "move the silver pot to the bottom left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0bba7c89149c614833390126.png"}, {"type": "image", "uri": "media/bdv2fail/0bba7c89149c614833390126.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0971/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 971, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "error_localization"}}
{"state": {"task": "move the stove below the right side burner", "subtask": "move the stove below the right side burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4e92b3270193e689d895bc7f.png"}, {"type": "image", "uri": "media/bdv2fail/2b7f1ea7785abb36b039438d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0972/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_76_train/7", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 972, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_76_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the silver pot below the white brick wall.", "subtask": "move the silver pot below the white brick wall.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9035bf4020ec0aeb2dda15ed.png"}, {"type": "image", "uri": "media/bdv2fail/4fa1fbd11168567b59be0351.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0974/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_39_train/10", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 974, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_39_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "fold the cloth from top to bottom", "subtask": "fold the cloth from top to bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d2090c366290ee5634ab062e.png"}, {"type": "image", "uri": "media/bdv2fail/d2090c366290ee5634ab062e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0982/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train/12", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 982, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train", "episode_id": "12", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "move the red pepper to the stove", "subtask": "move the red pepper to the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91eb7e3f7d640058d4a34155.png"}, {"type": "image", "uri": "media/bdv2fail/393f97f8699db3b519d02878.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0983/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_90_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 983, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_90_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "error_localization"}}
{"state": {"task": "put spatula beside but not on plate sink", "subtask": "put spatula beside but not on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f10a3cb9e75f3eacd9db109f.png"}, {"type": "image", "uri": "media/bdv2fail/d4d70bbb1008d410fc12937a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0984/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/40", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 984, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "40", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "put the plastic cup on the drying rack", "subtask": "put the plastic cup on the drying rack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b62905c73e12d56e18b4f0e8.png"}, {"type": "image", "uri": "media/bdv2fail/ea91fa5ba5f83870104327cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0986/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train/6", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 986, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "put spatula on yellow toy", "subtask": "put spatula on yellow toy", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50f930b58ce28cc7a4868de5.png"}, {"type": "image", "uri": "media/bdv2fail/9ee6e7f6f8355b0bcfd5698b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0987/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 987, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a57b7b0339419b130f4f705.png"}, {"type": "image", "uri": "media/bdv2fail/0a57b7b0339419b130f4f705.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0989/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/29", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 989, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "29", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "move the plastic yellow object to the left burner.", "subtask": "move the plastic yellow object to the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a2ab8f7cc079a4965de650f0.png"}, {"type": "image", "uri": "media/bdv2fail/b4aa86fc696697b9d4a98f84.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0990/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/13", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 990, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "put the white object inside the green object", "subtask": "put the white object inside the green object", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed6df83cafb25dade87dc4a8.png"}, {"type": "image", "uri": "media/bdv2fail/5c836022b9d3f9b4368c292a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0994/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train/14", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 994, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the yellow sponge to the right top burner.", "subtask": "move the yellow sponge to the right top burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "bdv2fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8fd7aa3066f81bb9a4d4371d.png"}, {"type": "image", "uri": "media/bdv2fail/2144839a7e0833f955030075.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0996/bdv2fail_exec_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/5", "family": "bdv2fail_exec_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 996, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "error_localization"}}
{"state": {"task": "put beet close to but not into pot sink", "subtask": "put beet close to but not into pot sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4a95b21b72bf1d1b6c929ed7.png"}, {"type": "image", "uri": "media/bdv2fail/8c3dd778407922c566c46d97.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0000/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 0, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the red knob to the bottom of the burner.", "subtask": "move the red knob to the bottom of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ea8a3c30cd86e1a31796e6f6.png"}, {"type": "image", "uri": "media/bdv2fail/1909b954247d2d67fa8fa5e1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0001/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 1, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2d7f63b399be4042896d73dc.png"}, {"type": "image", "uri": "media/bdv2fail/55a9a9eefb45666127278771.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0002/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 2, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the yellow knife underneath of the yellow cloth", "subtask": "put the yellow knife underneath of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/16f82423b82e81746422b57d.png"}, {"type": "image", "uri": "media/bdv2fail/6275ca7f7387deadf13e9b0f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0003/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 3, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "fold the cloth from the left to the right", "subtask": "fold the cloth from the left to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a68490151cc5678ddb2d25a3.png"}, {"type": "image", "uri": "media/bdv2fail/3d586d39805f107e817ddc20.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0004/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_12_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 4, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_12_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move an orange object to the top left of the table", "subtask": "move an orange object to the top left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70486cb351724ea807acf26b.png"}, {"type": "image", "uri": "media/bdv2fail/75e1f926e72ae04b4d3642c1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0005/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_11_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 5, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_11_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "place the brush near but not on the yellow cloth", "subtask": "place the brush near but not on the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/37c4dca1fb8c1328673f0976.png"}, {"type": "image", "uri": "media/bdv2fail/81860e40dd2ca4ba1b54c7a1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0006/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_10_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 6, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_10_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5a3f2d55fd39673d4e40df96.png"}, {"type": "image", "uri": "media/bdv2fail/9fbbca24a8a9fbb667d1f19c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0007/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 7, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "move silver pot to the lower side of right burner", "subtask": "move silver pot to the lower side of right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac0da747917faa49255523b8.png"}, {"type": "image", "uri": "media/bdv2fail/b5e5f684d1873562332badc4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0008/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_19_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 8, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_19_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the pot to the right burner", "subtask": "move the pot to the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/570171b69a2550d9fc082a3b.png"}, {"type": "image", "uri": "media/bdv2fail/a8a765d5b203941192c226af.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0009/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 9, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "place the blue cloth at the right end of the stove", "subtask": "place the blue cloth at the right end of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f03a19b48702438d7948fccb.png"}, {"type": "image", "uri": "media/bdv2fail/d7f84e26bf86fac5124f4b6f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0010/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_68_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 10, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_68_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the red fork to right corner of the burner", "subtask": "move the red fork to right corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b17a6539084af475afda36b0.png"}, {"type": "image", "uri": "media/bdv2fail/57a9c5e1cc1b031a55687035.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0011/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_40_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 11, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_40_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "unfold the cloth from the left to the right", "subtask": "unfold the cloth from the left to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/361c474fd33393a19ec2dda6.png"}, {"type": "image", "uri": "media/bdv2fail/3ea6bce91277201246871b9e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0012/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 12, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "take the silver pot to the lower left side of the table from the upper center", "subtask": "take the silver pot to the lower left side of the table from the upper center", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e347eb1a228e05441507ea5c.png"}, {"type": "image", "uri": "media/bdv2fail/5637b2725d4a5353bc7d991c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0013/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 13, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "fold the cloth from right to left", "subtask": "fold the cloth from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/320f052d6aead4bd61535fb4.png"}, {"type": "image", "uri": "media/bdv2fail/1d17876a509877d631eb6869.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0014/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_24_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 14, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_24_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "remove the red cuboid object from the right side of the table", "subtask": "remove the red cuboid object from the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/109070970ba60ad8fae0729f.png"}, {"type": "image", "uri": "media/bdv2fail/68cd038d94edc6d07fb40bd4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0015/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 15, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the banana to the lower right of the table", "subtask": "move the banana to the lower right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86753477bfea123aa804406e.png"}, {"type": "image", "uri": "media/bdv2fail/86753477bfea123aa804406e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0016/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 16, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the green cloth to the edge of the table", "subtask": "move the green cloth to the edge of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6060f3e8ccc37d9355d1b79.png"}, {"type": "image", "uri": "media/bdv2fail/dbaab0384a8b1fdc1eb9c05e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0017/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 17, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train", "episode_id": "18", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the yellow knife on top of the yellow cloth", "subtask": "put the yellow knife on top of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/16f82423b82e81746422b57d.png"}, {"type": "image", "uri": "media/bdv2fail/6275ca7f7387deadf13e9b0f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0018/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 18, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "turn faucet front to left.", "subtask": "turn faucet front to left.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9c3fd58e001fc1ff13d224aa.png"}, {"type": "image", "uri": "media/bdv2fail/a09d975c71db0eb7da9e95ee.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0019/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 19, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the silver pot and put it at the lower right of the right burner", "subtask": "move the silver pot and put it at the lower right of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/900cda0672f9d30f9fb96d2d.png"}, {"type": "image", "uri": "media/bdv2fail/abf4c160195bfe056dfc81d8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0020/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 20, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "remove the eggplant from the drawer", "subtask": "remove the eggplant from the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/baaae7251684e6571a4a277c.png"}, {"type": "image", "uri": "media/bdv2fail/b7c699726c0764aa80b18728.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0021/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_05_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 21, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_05_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upper left side of the table", "subtask": "move the silver pot to the upper left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/99672f5be4f08bd9cc9845ed.png"}, {"type": "image", "uri": "media/bdv2fail/a373a28da82e699e95c2c353.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0022/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 22, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the corn in between the burners", "subtask": "put the corn in between the burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2582749ce00c830a17349954.png"}, {"type": "image", "uri": "media/bdv2fail/8acbf18fd640c27d7a64daa7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0023/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 23, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the red tomato on plate", "subtask": "put the red tomato on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/521abf6e09490588cc42d896.png"}, {"type": "image", "uri": "media/bdv2fail/df54d632021295075ba06308.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0024/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 24, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "remove the yellow scrubber from the green cloth.", "subtask": "remove the yellow scrubber from the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c99f5cb2750d538b55f158b.png"}, {"type": "image", "uri": "media/bdv2fail/def3e801c49976c0d1f75979.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0025/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 25, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the silver pot on the orange towel", "subtask": "put the silver pot on the orange towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/90203c75a329c968da463e4a.png"}, {"type": "image", "uri": "media/bdv2fail/714039a9f66790bc1d9bce11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0026/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 26, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the blue brush to the upper left corner of the stove", "subtask": "move the blue brush to the upper left corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ee681ef9d8cb6bc96f60e4a5.png"}, {"type": "image", "uri": "media/bdv2fail/879ee3e55cc09c57871e37a8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0027/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 27, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upper left of the stove", "subtask": "move the silver pot to the upper left of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/54a9552c030c81ceb14d760b.png"}, {"type": "image", "uri": "media/bdv2fail/856bd616d97ff1df4d7124d5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0028/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 28, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth to the left of the table", "subtask": "move the cloth to the left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/20030bbeac881a51e0322251.png"}, {"type": "image", "uri": "media/bdv2fail/54d4853d62d68aca90950647.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0029/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_04_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 29, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_04_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the right side of the burner", "subtask": "move the yellow cloth to the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4a39f48213acb40ccf4b537e.png"}, {"type": "image", "uri": "media/bdv2fail/4a39f48213acb40ccf4b537e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0030/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 30, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "3", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue plastic spoon to the left side of the blue fork.", "subtask": "move the blue plastic spoon to the left side of the blue fork.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/98d58ec9799f25aac67d0e37.png"}, {"type": "image", "uri": "media/bdv2fail/886314127642897041dfca29.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0031/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 31, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "3", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the upper left of the red pepper", "subtask": "move the blue spoon to the upper left of the red pepper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/681f3d016aa9346d987aade2.png"}, {"type": "image", "uri": "media/bdv2fail/250740c61b1792ea2e94b3be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0032/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 32, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "place the metal pot on the front left edge of the stove.", "subtask": "place the metal pot on the front left edge of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a49ff3162c5c99ba207f9477.png"}, {"type": "image", "uri": "media/bdv2fail/9158e2ea2fd8b8337b56cea7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0033/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_30_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 33, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_30_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the green bell pepper in front of the drawer.", "subtask": "move the green bell pepper in front of the drawer.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9453cabac20bab1bc926b8dd.png"}, {"type": "image", "uri": "media/bdv2fail/95219641a8a28212a302ef30.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0034/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/34", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 34, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "34", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "fold the cloth from the bottom to the top", "subtask": "fold the cloth from the bottom to the top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1e7dc589b8d8a9b991986e3.png"}, {"type": "image", "uri": "media/bdv2fail/22d23e8848f108cd96ff6400.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0035/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_05_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 35, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_05_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove the silver pan from the purple rag.", "subtask": "remove the silver pan from the purple rag.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/751f8cdfe4fb4cc7de9e2264.png"}, {"type": "image", "uri": "media/bdv2fail/785c3ec134714c596aba4e43.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0036/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 36, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the red object to the bottom right of the stove", "subtask": "move the red object to the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ca03d1ed51d690c9da373fb.png"}, {"type": "image", "uri": "media/bdv2fail/9ca03d1ed51d690c9da373fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0037/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 37, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the black cord from right to left.", "subtask": "move the black cord from right to left.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4798d42389e06450c4a519dd.png"}, {"type": "image", "uri": "media/bdv2fail/e3947a77050cb45c5876be82.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0038/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 38, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "unfold the cloth from left to right side of the table", "subtask": "unfold the cloth from left to right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/10e83a34920428c4f7de9197.png"}, {"type": "image", "uri": "media/bdv2fail/d377ce3be82a28180de0f7eb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0039/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_19_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 39, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_19_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the spoon in the green pepper", "subtask": "put the spoon in the green pepper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/416a77f5037762a3a284e9b7.png"}, {"type": "image", "uri": "media/bdv2fail/6818c3cb8b055b233438070d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0040/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 40, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the cloth to the left burner from the right burner", "subtask": "move the cloth to the left burner from the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e9c678dba3c0355a6ece7fa3.png"}, {"type": "image", "uri": "media/bdv2fail/0475b18118c518a6e66c64b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0041/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 41, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move blue ladle 3 inches to the right", "subtask": "move blue ladle 3 inches to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a9f945518296e4eca72ae392.png"}, {"type": "image", "uri": "media/bdv2fail/0de36e81b52167b254056733.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0042/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_99_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 42, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_99_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the red fork on the upper left side of the table over the pot", "subtask": "move the red fork on the upper left side of the table over the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/18be481c9b51e172248e495d.png"}, {"type": "image", "uri": "media/bdv2fail/e17378ae2f5a1d7934a985e6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0043/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 43, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the yellow knife to top rigth of burner", "subtask": "move the yellow knife to top rigth of burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c67cca250de9fd7b64d2f5dd.png"}, {"type": "image", "uri": "media/bdv2fail/a307de57467c1d3ddf3a6255.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0044/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_08_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 44, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_08_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the little box to the top right side of the stove", "subtask": "move the little box to the top right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1af61238df201834801aaf5f.png"}, {"type": "image", "uri": "media/bdv2fail/b7736924a83e75ea82de41f8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0045/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 45, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the silver pot in the right side of the mushroom", "subtask": "move the silver pot in the right side of the mushroom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/96aadcbe4b5f5e60bf978cdb.png"}, {"type": "image", "uri": "media/bdv2fail/426ceddab3b5b256bbe333c0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0046/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 46, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot and put it on the violet cloth", "subtask": "move the silver pot and put it on the violet cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/82e0094d83ecf30fef7e5d22.png"}, {"type": "image", "uri": "media/bdv2fail/4f104a265d3a82f8651d0890.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0047/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_37_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 47, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_37_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "place the yellow knife to the upper side of the blue dish towel.", "subtask": "place the yellow knife to the upper side of the blue dish towel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2d644706ba3f43a9678efd1e.png"}, {"type": "image", "uri": "media/bdv2fail/e275bfe7b6933a9d765a5d5e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0048/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_62_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 48, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_62_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "place the red spoon on the yellow cloth", "subtask": "place the red spoon on the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0bcbf6d75dbd334f7d462e82.png"}, {"type": "image", "uri": "media/bdv2fail/d1ff577c463d567c07e66568.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0049/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_109_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 49, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_109_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the silver frying pan to the right of the green towel.", "subtask": "move the silver frying pan to the right of the green towel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d48b0b79c11fbd1ce0ebbfb9.png"}, {"type": "image", "uri": "media/bdv2fail/a33bdde5f4ed17c5d274427c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0050/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 50, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the yellow knife to the left side of the left burner", "subtask": "move the yellow knife to the left side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c835af7b3f50870e2880974.png"}, {"type": "image", "uri": "media/bdv2fail/de985a24f417c71fc07959d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0051/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 51, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the blue object to the left edge of the table from the bottom right edge of the orange cloth.", "subtask": "move the blue object to the left edge of the table from the bottom right edge of the orange cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f0f4ca89a4eac09cb8d16a8.png"}, {"type": "image", "uri": "media/bdv2fail/acf4c124cb007ee22fb62960.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0052/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 52, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 253, "role": "outcome"}}
{"state": {"task": "move the spatula to the right side of the silver pot from the upper part of the table", "subtask": "move the spatula to the right side of the silver pot from the upper part of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6ff11b85fb34c61b405aed0.png"}, {"type": "image", "uri": "media/bdv2fail/5311243f1d47f128a0f673d2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0053/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_02_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 53, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_02_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to bottom of the left burner", "subtask": "move the yellow cloth to bottom of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/36e6dea4c101446f80bbd987.png"}, {"type": "image", "uri": "media/bdv2fail/40b4d8787627182edb57e254.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0054/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_31_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 54, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_31_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the red cup on the green pepper", "subtask": "put the red cup on the green pepper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a02ccaed04137e760423041f.png"}, {"type": "image", "uri": "media/bdv2fail/0962ac2f872e8c47534f515d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0055/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 55, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the cloth to top edge on the table", "subtask": "move the cloth to top edge on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/739dd9c3f39371be25423387.png"}, {"type": "image", "uri": "media/bdv2fail/49fb1f32a27121af2fe97364.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0056/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 56, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the banana stuff toy above the tiger stuff toy", "subtask": "move the banana stuff toy above the tiger stuff toy", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5a360debf4ccdc929ea5d8d7.png"}, {"type": "image", "uri": "media/bdv2fail/7adc7b8f9dedfd71a22db8c0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0057/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_15_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 57, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_15_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the top left side of the table.", "subtask": "remove the silver pot from the top left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec81563932a7af2df0c4108e.png"}, {"type": "image", "uri": "media/bdv2fail/995aff5f8aec2739c6e6659e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0058/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 58, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the silver pan to the left side of the stove", "subtask": "move the silver pan to the left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/060fb26aa869ed92663da79c.png"}, {"type": "image", "uri": "media/bdv2fail/19cb7c8614612be205f3df82.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0059/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_41_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 59, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_41_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth from left to right", "subtask": "move the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d5da668f57dbae50dbca495.png"}, {"type": "image", "uri": "media/bdv2fail/9fd3b47bcf9d897c052e765e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0060/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 60, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "fold the cloth from the bottom to the top", "subtask": "fold the cloth from the bottom to the top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/44e6760789274d994cc4f729.png"}, {"type": "image", "uri": "media/bdv2fail/db16e5aefc47ba14f0e1fbd5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0061/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 61, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom right side of the stove", "subtask": "move the silver pot to the bottom right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/34cd516bb80990b2213577f0.png"}, {"type": "image", "uri": "media/bdv2fail/34cd516bb80990b2213577f0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0062/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 62, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "18", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the left burner", "subtask": "move the yellow cloth to the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b0d33a31869e24098ea87ccb.png"}, {"type": "image", "uri": "media/bdv2fail/0ef136ad2dd61cef9aa9c5c0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0063/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 63, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the blue spatula to the bottom of burner.", "subtask": "put the blue spatula to the bottom of burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/63e2064e24694a579586d64f.png"}, {"type": "image", "uri": "media/bdv2fail/d6f574863a9581575e406025.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0064/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_45_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 64, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_45_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "unfold the cloth from center to top right", "subtask": "unfold the cloth from center to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2607827e9d2281c2f1e93494.png"}, {"type": "image", "uri": "media/bdv2fail/24446ba1b3b8177dd96906a9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0065/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_06_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 65, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_06_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the red object to bottom left edge of the cloth", "subtask": "place the red object to bottom left edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d97be847a74a4791e62d940.png"}, {"type": "image", "uri": "media/bdv2fail/a1fca23ea9e591b9384dd290.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0066/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 66, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc373be66bb13a2241110e5c.png"}, {"type": "image", "uri": "media/bdv2fail/cc373be66bb13a2241110e5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0067/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_06_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 67, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_06_train", "episode_id": "21", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "slide the towel towards the left side of the stove.", "subtask": "slide the towel towards the left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/159ccec4ee3c0e1b871b8ec3.png"}, {"type": "image", "uri": "media/bdv2fail/37589c6c3f1dd7c3fd35914c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0068/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 68, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "unfold cloth top left to bottom right", "subtask": "unfold cloth top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/afde9f6fc07fdaeb80746652.png"}, {"type": "image", "uri": "media/bdv2fail/b553f1a2cb6cc9f8fdfb17a1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0069/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 69, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "place the purple figure at the top left of the cloth", "subtask": "place the purple figure at the top left of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/07af6127d9cb20e9964817d8.png"}, {"type": "image", "uri": "media/bdv2fail/dde9e92b963f81b80104fd8f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0070/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_12_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 70, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_12_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot next to the light green handle spatula.", "subtask": "move the silver pot next to the light green handle spatula.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c294208b6fde9d27d9506dc1.png"}, {"type": "image", "uri": "media/bdv2fail/08ff4d2f3451d36b3ba5cc03.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0071/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 71, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train", "episode_id": "20", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the banana to the upper middle of the toy stove.", "subtask": "move the banana to the upper middle of the toy stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2a019097dd9a28e6ed69d6f9.png"}, {"type": "image", "uri": "media/bdv2fail/3226f8e26736e315615ac3c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0072/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 72, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the yellow knife on the right burner", "subtask": "put the yellow knife on the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/699326b30b4915fbdcf6f292.png"}, {"type": "image", "uri": "media/bdv2fail/2823dcc99b327a3b4448f98d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0073/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_02_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 73, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_02_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the banana from the napkin from the left side of the table", "subtask": "remove the banana from the napkin from the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f6334491f709ce9f247bad63.png"}, {"type": "image", "uri": "media/bdv2fail/91d129be15c2df2fd7b2cbe9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0074/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 74, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom left corner of the burner", "subtask": "move the silver pot to the bottom left corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c90750c2ccb5c44fcd5f9de5.png"}, {"type": "image", "uri": "media/bdv2fail/d8710242fa889b3b3f52320a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0075/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 75, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth from bottom to top", "subtask": "move the cloth from bottom to top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c46bd3da739ec4bfb2c2403f.png"}, {"type": "image", "uri": "media/bdv2fail/1af6cec5704137455abc579e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0076/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 76, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put red pepper on plate", "subtask": "put red pepper on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6ad1a3c7322d7f9b7d118cd.png"}, {"type": "image", "uri": "media/bdv2fail/7a0a8de7f4dbb945917e87e9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0077/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 77, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the yellow knife to the upper side of the blue cloth.", "subtask": "move the yellow knife to the upper side of the blue cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9ce29d894602fc9b44bea0e.png"}, {"type": "image", "uri": "media/bdv2fail/714d552e8f8084419258beb6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0078/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_108_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 78, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_108_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to top right", "subtask": "fold the cloth from bottom right to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a537636f82d031a2a44d87e9.png"}, {"type": "image", "uri": "media/bdv2fail/7c6918fd9ae4aefae5cd9c16.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0079/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 79, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put cup to counter from sink", "subtask": "put cup to counter from sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f98837c6ee86f5a7418ac017.png"}, {"type": "image", "uri": "media/bdv2fail/7dc71afe9e563f71321cd84e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0080/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/51", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 80, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "51", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "moving the green spatula into the right side of the stove", "subtask": "moving the green spatula into the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf70ef1fb3814e12ee7d1a4f.png"}, {"type": "image", "uri": "media/bdv2fail/111b934ecc306d90bc23feec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0081/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 81, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put cup to counter from sink", "subtask": "put cup to counter from sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/56410136666ac62ea65455bb.png"}, {"type": "image", "uri": "media/bdv2fail/01f29298441c27535fe96ec6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0082/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/66", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 82, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "66", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "remove corn from bowl sink", "subtask": "remove corn from bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ce6fc7453df675c8dc87c881.png"}, {"type": "image", "uri": "media/bdv2fail/1da9113e50b0c1b99eebec00.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0083/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 83, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put cup on plate", "subtask": "put cup on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7ef2a9e672d01087d4708ce4.png"}, {"type": "image", "uri": "media/bdv2fail/7ef2a9e672d01087d4708ce4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0084/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/27", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 84, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "27", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "turn faucet front to left.", "subtask": "turn faucet front to left.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/03b196810f1e2d3e9a3cb012.png"}, {"type": "image", "uri": "media/bdv2fail/65048c1595538fd5727b45b5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0085/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 85, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the yellow spoon to the top left side of the table", "subtask": "move the yellow spoon to the top left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c8686b03466f9788174cac01.png"}, {"type": "image", "uri": "media/bdv2fail/c5f1e47d432106bf606a48e4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0086/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 86, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth to the left from the right", "subtask": "move the cloth to the left from the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3337c2945481b99ef0c3388d.png"}, {"type": "image", "uri": "media/bdv2fail/074338752e9d350bb97a6c61.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0087/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_25_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 87, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_25_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the silver pot from the top right corner of the cook to the left burner.", "subtask": "move the silver pot from the top right corner of the cook to the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/051f24ca4e6f0d672d3560ca.png"}, {"type": "image", "uri": "media/bdv2fail/aca3a0b630d2b7fd14bfef3e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0088/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 88, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "put the green object in between the burners", "subtask": "put the green object in between the burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4585d0acc38272f46e71d7c4.png"}, {"type": "image", "uri": "media/bdv2fail/6ea37b21854ef0c98071472d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0089/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 89, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the pot to the front of the purple cloth.", "subtask": "move the pot to the front of the purple cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/012737e5a059e96082b64d7b.png"}, {"type": "image", "uri": "media/bdv2fail/c0ff59fb4cf9a676c0751506.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0090/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_101_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 90, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_101_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot to the the bottom right corner of the stove", "subtask": "move the silver pot to the the bottom right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc22283f05ca1cb70f574f5e.png"}, {"type": "image", "uri": "media/bdv2fail/216f528e5995225a23f24da7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0091/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_74_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 91, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_74_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top left to left", "subtask": "unfold the cloth from top left to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f1bfee9fd3421583484942e5.png"}, {"type": "image", "uri": "media/bdv2fail/c17ffe67170ac6fc9dc1ca86.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0092/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_07_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 92, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_07_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the blue fork to the upper right side of the stove next to the red object", "subtask": "move the blue fork to the upper right side of the stove next to the red object", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d361addd4233dde18166a4f.png"}, {"type": "image", "uri": "media/bdv2fail/aba1bcdd35abae2c307638e9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0093/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_107_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 93, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_107_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "moves the fruit from the cloth to the left of the table", "subtask": "moves the fruit from the cloth to the left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/715ae8662b89c1c33922f1d0.png"}, {"type": "image", "uri": "media/bdv2fail/78ed34c6cfdc5badde02afe8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0094/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train/42", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 94, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train", "episode_id": "42", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the silver pan from the upper right side of stove", "subtask": "remove the silver pan from the upper right side of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2f947d4d4c049667755a07e7.png"}, {"type": "image", "uri": "media/bdv2fail/112623062501ef837e66e25a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0095/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 95, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/54b3bd170b6e8e6d4263f201.png"}, {"type": "image", "uri": "media/bdv2fail/f9710df9e5c7b959b6f5f6c2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0096/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 96, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "move the red spoon to the front right corner of the stove", "subtask": "move the red spoon to the front right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed39746680a92147fbfe4673.png"}, {"type": "image", "uri": "media/bdv2fail/612ffba806eb0336038c68fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0097/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 97, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the ladle spoon from top of the yellow cloth", "subtask": "remove the ladle spoon from top of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f10af344b848619c590dd50f.png"}, {"type": "image", "uri": "media/bdv2fail/e7593a242e11348a1974bc1e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0098/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_47_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 98, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_47_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow cloth upward the stove", "subtask": "move the yellow cloth upward the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52938137d441f4189764d1d7.png"}, {"type": "image", "uri": "media/bdv2fail/7cbaffda30e75e1560a38b82.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0099/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 99, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the purple cloth to the upper left of the table", "subtask": "move the purple cloth to the upper left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/78281ccf49ecd2d72f02cc35.png"}, {"type": "image", "uri": "media/bdv2fail/78281ccf49ecd2d72f02cc35.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0100/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 100, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train", "episode_id": "12", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot on top of the toy mushroom.", "subtask": "move the silver pot on top of the toy mushroom.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dfd421c8d0c3cbba5e2cc99.png"}, {"type": "image", "uri": "media/bdv2fail/9c0371fec49a295951bdf16a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0101/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 101, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the bottom right of the stove", "subtask": "move the blue spoon to the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/355752f3145a544ae2844058.png"}, {"type": "image", "uri": "media/bdv2fail/42e2bbe705ed69934072aa19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0102/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 102, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the metal pot to the left of the stove.", "subtask": "move the metal pot to the left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ce978411a6b19e94fbcaadd.png"}, {"type": "image", "uri": "media/bdv2fail/684f90ef6eaf3a4dfb791409.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0103/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 103, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "remove the corn from the upper left side of the burner", "subtask": "remove the corn from the upper left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e6fd821f7170f246601b6de3.png"}, {"type": "image", "uri": "media/bdv2fail/05f00798810589c29fe22dd1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0104/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_79_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 104, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_79_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the red spoon to the bottom left corner of the burner.", "subtask": "move the red spoon to the bottom left corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bfe26fc80a74a4ab7a70bf8b.png"}, {"type": "image", "uri": "media/bdv2fail/299d06fc30b1437f3087f8a8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0105/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_35_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 105, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_35_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the yellow knife from inbetween the left burner and the can and put it to bottom of left burner", "subtask": "move the yellow knife from inbetween the left burner and the can and put it to bottom of left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b7f4b69e7d09ab12b82766d8.png"}, {"type": "image", "uri": "media/bdv2fail/2a8be9055f21608e88c47aff.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0106/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 106, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 255, "role": "outcome"}}
{"state": {"task": "place the tall can in the middle of the left heating plate", "subtask": "place the tall can in the middle of the left heating plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6d18a7a74a8d0f008df73117.png"}, {"type": "image", "uri": "media/bdv2fail/89840140bf26954e1e2e1b3c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0107/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 107, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red spoon from the back left to the front right corner", "subtask": "move the red spoon from the back left to the front right corner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bffc7ba1b7cc212d6ed0a846.png"}, {"type": "image", "uri": "media/bdv2fail/ec02bd65b6583fd4c5d4b94a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0108/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 108, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "put the the fork to the top of the pot.", "subtask": "put the the fork to the top of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c83329436a8a36524b8ae752.png"}, {"type": "image", "uri": "media/bdv2fail/fd60c5307251f43e9854be6c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0109/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_43_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 109, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_43_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the green object above the left burner", "subtask": "move the green object above the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f0afd06e24fb1e558da52247.png"}, {"type": "image", "uri": "media/bdv2fail/3eda4883112270cc3a63e278.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0110/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 110, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the yellow duck above the drawer", "subtask": "put the yellow duck above the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/41fcb48a11b60f5deb2b499d.png"}, {"type": "image", "uri": "media/bdv2fail/9e8f843e899beaa3131cc0e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0111/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/43", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 111, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "43", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "remove the plastic ice from the bottom edge of the cloth", "subtask": "remove the plastic ice from the bottom edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/09c9c925bf56123519da8e19.png"}, {"type": "image", "uri": "media/bdv2fail/a7c05b89a50defa1b0ec5d34.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0112/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 112, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the red cup below the right stove", "subtask": "put the red cup below the right stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a85555d4fbc032e2c5db5b3e.png"}, {"type": "image", "uri": "media/bdv2fail/8aed990a0570d97095766dec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0113/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 113, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the spatula and place it just behind the pan", "subtask": "move the spatula and place it just behind the pan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79e26cf73d134fe9b9181273.png"}, {"type": "image", "uri": "media/bdv2fail/d7260464af59a8e5e633d2cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0114/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 114, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot to the left of the stove", "subtask": "move the silver pot to the left of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a1b6bdcf5c4abae1aa33d060.png"}, {"type": "image", "uri": "media/bdv2fail/835737d5e4143a5de5b5c503.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0115/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 115, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the cloth from the left bottom side of the cooker to the top right side", "subtask": "move the cloth from the left bottom side of the cooker to the top right side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a440eaafa8ed44241d69ebf8.png"}, {"type": "image", "uri": "media/bdv2fail/6fe0c0bb254d0d639d9420bd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0116/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 116, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move the yellow towel beside the yellow microwave.", "subtask": "move the yellow towel beside the yellow microwave.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/99deef4da2a4180d5fbcc114.png"}, {"type": "image", "uri": "media/bdv2fail/333f6261cb1ce832a08d253b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0117/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 117, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the yellow cloth", "subtask": "remove the silver pot from the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0c7e445be1566630533750f6.png"}, {"type": "image", "uri": "media/bdv2fail/439f1fdf11263d35555de211.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0118/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 118, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move silver pot to left burner", "subtask": "move silver pot to left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9fb2ff2798adba1910cd2bb1.png"}, {"type": "image", "uri": "media/bdv2fail/e8d13df543b05c20b0fa5173.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0119/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_71_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 119, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_71_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the can to the lower left corner of the stove", "subtask": "move the can to the lower left corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/344616d784774ec413d342e1.png"}, {"type": "image", "uri": "media/bdv2fail/344616d784774ec413d342e1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0120/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 120, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "0", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "unfold the cloth from bottom right to top left", "subtask": "unfold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2e4087e43ae183479618bfeb.png"}, {"type": "image", "uri": "media/bdv2fail/3beed578dadec84fbaace477.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0121/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/32", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 121, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "32", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the brush to the upper right corner of the cooker", "subtask": "move the brush to the upper right corner of the cooker", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/45a2bb8bd12c8c54b1f02822.png"}, {"type": "image", "uri": "media/bdv2fail/9d4d5ffa08d28f97c0d6f272.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0122/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 122, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot in the bottom right side of the burner", "subtask": "move the silver pot in the bottom right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0ddc5fe40decc5c40c65f500.png"}, {"type": "image", "uri": "media/bdv2fail/bc8ad945e340138b4bd66b25.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0123/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 123, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "slide purple cloth over to the front right corner of the stove", "subtask": "slide purple cloth over to the front right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2cc02ef2b86dd5c8959b69ac.png"}, {"type": "image", "uri": "media/bdv2fail/499bdb481bd0fa60997184ad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0124/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 124, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth from right to left", "subtask": "move the cloth from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4798d42389e06450c4a519dd.png"}, {"type": "image", "uri": "media/bdv2fail/e3947a77050cb45c5876be82.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0125/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 125, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move silver pot to the top corner of the left side of the burner", "subtask": "move silver pot to the top corner of the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed600c58effec74a1024c79c.png"}, {"type": "image", "uri": "media/bdv2fail/88fd644504c0d1f258062405.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0126/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_70_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 126, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_70_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the yellow triangle from left to top of the table.", "subtask": "move the yellow triangle from left to top of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d0f2b998594da7d0f35bfba.png"}, {"type": "image", "uri": "media/bdv2fail/ec36e43732920ab5d43c21c4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0127/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_01_train/28", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 127, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_01_train", "episode_id": "28", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth from right to left edge of the sink", "subtask": "move the cloth from right to left edge of the sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/175579bbb75198f80dce1115.png"}, {"type": "image", "uri": "media/bdv2fail/419b835930a7f158e21f582e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0128/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 128, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "pick up blue pen and put next to but not into drawer", "subtask": "pick up blue pen and put next to but not into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4233b866de77b685595cea8.png"}, {"type": "image", "uri": "media/bdv2fail/dbb82d5ec344c72ef5bdee42.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0129/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 129, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red spoon to the upper right corner of the stove.", "subtask": "move the red spoon to the upper right corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/87ebcea7899966035986f523.png"}, {"type": "image", "uri": "media/bdv2fail/c47de527fc3df30fd301238a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0130/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_28_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 130, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_28_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the green object under the yellow towel.", "subtask": "put the green object under the yellow towel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9cbda08bed65dbad9149135.png"}, {"type": "image", "uri": "media/bdv2fail/fac9757433446bba7912b7f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0131/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_42_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 131, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_42_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the blue spoon from the top of the green towel", "subtask": "remove the blue spoon from the top of the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5e256504f59abbfbc808e14.png"}, {"type": "image", "uri": "media/bdv2fail/b5e256504f59abbfbc808e14.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0132/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_31_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 132, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_31_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the green sponge to right of stove from left burner", "subtask": "move the green sponge to right of stove from left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0812e9a4727a17e46891ba5f.png"}, {"type": "image", "uri": "media/bdv2fail/7921bf8c20b4ffe262d88957.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0133/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 133, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the blue figure on the top edge of the cloth", "subtask": "put the blue figure on the top edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1120deb33af8bd59b4f33f0f.png"}, {"type": "image", "uri": "media/bdv2fail/c7a289a37865edbe0b98f039.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0134/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_23_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 134, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_23_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue brush to the bottom right of the stove", "subtask": "move the blue brush to the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f1289283fafebd3eff364ef6.png"}, {"type": "image", "uri": "media/bdv2fail/772e827220c2ad2eb399b8d8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0135/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_09_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 135, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_09_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top left to bottom right", "subtask": "unfold the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7ba63ab5b49a1b9ff96e47a5.png"}, {"type": "image", "uri": "media/bdv2fail/87240fb667a68b5b82c8b930.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0136/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_17_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 136, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_17_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the yellow towel from the right lower part of the cooker to the right burner", "subtask": "move the yellow towel from the right lower part of the cooker to the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eaa83a81d47c066a43e64768.png"}, {"type": "image", "uri": "media/bdv2fail/e21d68565a6a055703b71c73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0137/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_42_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 137, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_42_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "put cucumber next to but not in cup", "subtask": "put cucumber next to but not in cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f5381e25124df9b002dcb3e1.png"}, {"type": "image", "uri": "media/bdv2fail/a88a871d37a7ec36a428f42e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0138/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/29", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 138, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "29", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the white object to the left burner", "subtask": "move the white object to the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5792dc5a2505cf8dbf06190.png"}, {"type": "image", "uri": "media/bdv2fail/c5792dc5a2505cf8dbf06190.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0139/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_15_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 139, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_15_train", "episode_id": "8", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the plastic ice on the bottom edge of the cloth", "subtask": "put the plastic ice on the bottom edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a7c05b89a50defa1b0ec5d34.png"}, {"type": "image", "uri": "media/bdv2fail/a7c05b89a50defa1b0ec5d34.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0140/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 140, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "18", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom left to the right", "subtask": "fold the cloth from bottom left to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6348b16ba720e945249a5418.png"}, {"type": "image", "uri": "media/bdv2fail/dfcd3f17aa9866ea13ee5bdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0141/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_12_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 141, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_12_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the blue strawberry from the right side of the table to the left.", "subtask": "move the blue strawberry from the right side of the table to the left.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c8dddd5afa781e2031e21f66.png"}, {"type": "image", "uri": "media/bdv2fail/360b11cc7d6327ef5326f027.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0142/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 142, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the blue brush to the upper right corner of the stove", "subtask": "move the blue brush to the upper right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ee681ef9d8cb6bc96f60e4a5.png"}, {"type": "image", "uri": "media/bdv2fail/879ee3e55cc09c57871e37a8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0143/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 143, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the blue spoon around but not on the top of the green towel", "subtask": "put the blue spoon around but not on the top of the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c728ac183e0042e2196d76b8.png"}, {"type": "image", "uri": "media/bdv2fail/b5e256504f59abbfbc808e14.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0144/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_31_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 144, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_31_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top left to bottom right", "subtask": "unfold the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f9aace0806cc7c58e7928e38.png"}, {"type": "image", "uri": "media/bdv2fail/68ecb734873601e36553fea1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0145/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_18_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 145, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_18_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "grab the cloth and move it to left from right of the table", "subtask": "grab the cloth and move it to left from right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3ba7a337fc936a4054f10472.png"}, {"type": "image", "uri": "media/bdv2fail/f131f931ddcf72d4b20af7bb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0146/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 146, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the can to the right side of the burner", "subtask": "move the can to the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4aa612033284980f7765c303.png"}, {"type": "image", "uri": "media/bdv2fail/03344384c86720dd51d5d26e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0147/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_02_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 147, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_02_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put spoon in bowl sink", "subtask": "put spoon in bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/538131472a947bbd1dc11768.png"}, {"type": "image", "uri": "media/bdv2fail/7082aef3fedea0554f870c19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0148/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 148, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "remove the orange fish from the silver bowl", "subtask": "remove the orange fish from the silver bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91da124d73ab01a9507065c6.png"}, {"type": "image", "uri": "media/bdv2fail/ccb141ee88df957f9caf0b52.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0149/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 149, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the green object inside the drawer", "subtask": "place the green object inside the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e91d0dd17e8e5cae5243902a.png"}, {"type": "image", "uri": "media/bdv2fail/6751c9bd7fa2124cc4ebd2fc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0150/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_01_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 150, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_01_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put cup to counter from sink", "subtask": "put cup to counter from sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c990e8c0e97d6dc50dfc9b5f.png"}, {"type": "image", "uri": "media/bdv2fail/bf99c287245707462d9c9a16.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0151/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 151, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "take cucumber out of cup", "subtask": "take cucumber out of cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/43d46f19be8083c91bfb2a71.png"}, {"type": "image", "uri": "media/bdv2fail/a7128612a085af4d0cf1b6d6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0152/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_take_cucumber_out_of_cup_train/41", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 152, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_take_cucumber_out_of_cup_train", "episode_id": "41", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "take banana out of colander", "subtask": "take banana out of colander", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/73558bd4324ba0852358873a.png"}, {"type": "image", "uri": "media/bdv2fail/9d63a63fc3edfd87269fd454.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0153/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 153, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/513b8cad55480d76db8a44ed.png"}, {"type": "image", "uri": "media/bdv2fail/513b8cad55480d76db8a44ed.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0154/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 154, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train", "episode_id": "12", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the red fork on top of the grey plastic sink", "subtask": "put the red fork on top of the grey plastic sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79bbdba096a58685f6e196c3.png"}, {"type": "image", "uri": "media/bdv2fail/cc04978cf358f532cd05cd53.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0155/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_43_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 155, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_43_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the pot lid on the blue pot holder", "subtask": "move the pot lid on the blue pot holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/83233fc8f60c7caa1b1fa065.png"}, {"type": "image", "uri": "media/bdv2fail/07ec64235cdca9c08d37ba49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0156/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_11_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 156, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_11_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "take green pepper and place it in the plated pot", "subtask": "take green pepper and place it in the plated pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eaad6b8b6b46802af4632e32.png"}, {"type": "image", "uri": "media/bdv2fail/ee9a55d254b950a83f17f313.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0157/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_09_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 157, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_09_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the orange cloth on top of the red object", "subtask": "put the orange cloth on top of the red object", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f7f2d6ef8aa1897c31a8dd0.png"}, {"type": "image", "uri": "media/bdv2fail/dfa272cabf1705b115676298.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0158/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 158, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the green spatula slighly to the left of the burner and place theblue thing below the burner", "subtask": "move the green spatula slighly to the left of the burner and place theblue thing below the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b474876d42b9d67e8e47bfbb.png"}, {"type": "image", "uri": "media/bdv2fail/18a30f73a9e0d9e6bef22fb1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0159/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 159, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 257, "role": "outcome"}}
{"state": {"task": "move the cloth to the left of the pot.", "subtask": "move the cloth to the left of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e518077a57f79b81c7097538.png"}, {"type": "image", "uri": "media/bdv2fail/dc0da7265154d808b6b8fd75.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0160/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_06_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 160, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_06_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom right corner of the stove", "subtask": "move the silver pot to the bottom right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/82d3f4f7b498331b26b02b32.png"}, {"type": "image", "uri": "media/bdv2fail/125ed60fb0177a3bb7115edd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0161/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_47_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 161, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_47_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red fork to the middle left side of the table", "subtask": "move the red fork to the middle left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8b52f940e139d6930d040091.png"}, {"type": "image", "uri": "media/bdv2fail/dd38e34630c4f7bd1788d3f5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0162/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_39_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 162, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_39_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove cucumber from cup", "subtask": "remove cucumber from cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/378ef75919fd6216f4a79b58.png"}, {"type": "image", "uri": "media/bdv2fail/2dc5266cb77299e2ccd10ce9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0163/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 163, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "place the red pepper close to but not on the blue towel", "subtask": "place the red pepper close to but not on the blue towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4063c0879df8e0e01c39798a.png"}, {"type": "image", "uri": "media/bdv2fail/8376c3ead33442a7aa79fd2f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0164/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_02_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 164, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_02_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "take blueberries off plate sink", "subtask": "take blueberries off plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7fe5424b9ea9ae3749059a79.png"}, {"type": "image", "uri": "media/bdv2fail/8a798df24a1eb8241633c827.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0165/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_blueberries_off_plate_sink_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 165, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_blueberries_off_plate_sink_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the silver pot lid from the pot to the yellow cloth on the top right of the table", "subtask": "move the silver pot lid from the pot to the yellow cloth on the top right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8ccfe7756ea7cb41987b5ec2.png"}, {"type": "image", "uri": "media/bdv2fail/9b31e86b8bb31773b6d23303.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0166/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 166, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "put the yellow object next to but not on th ecloth in", "subtask": "put the yellow object next to but not on th ecloth in", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a48e1e2b8bbafae842bc5620.png"}, {"type": "image", "uri": "media/bdv2fail/18ce7a1e0faa7922f55bf051.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0167/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 167, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the green cloth to the bottom of the left burner", "subtask": "move the green cloth to the bottom of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2ea6339a920c791e06c9e5f2.png"}, {"type": "image", "uri": "media/bdv2fail/2ea6339a920c791e06c9e5f2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0168/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_40_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 168, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_40_train", "episode_id": "5", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove the grapes from the table and put it in the pan", "subtask": "remove the grapes from the table and put it in the pan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8bc3cfef84c83d95dc48e8c2.png"}, {"type": "image", "uri": "media/bdv2fail/8bc3cfef84c83d95dc48e8c2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0169/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_00_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 169, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_00_train", "episode_id": "3", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the green object from between the burners", "subtask": "remove the green object from between the burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6ea37b21854ef0c98071472d.png"}, {"type": "image", "uri": "media/bdv2fail/4585d0acc38272f46e71d7c4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0170/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 170, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the silver pot into the upper right side of the table", "subtask": "move the silver pot into the upper right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ea0260f4e1820fd788040e2a.png"}, {"type": "image", "uri": "media/bdv2fail/6fd35254cc36ac6391e44cb6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0171/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 171, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "the robot bend cloth", "subtask": "the robot bend cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4bacdc0ac506d38338d1c492.png"}, {"type": "image", "uri": "media/bdv2fail/783bc6c8089ad0f33dab0e31.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0172/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train/35", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 172, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train", "episode_id": "35", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the orange cloth to the bottom of right burner.", "subtask": "move the orange cloth to the bottom of right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/87bfe40c95d7e341ca326fa5.png"}, {"type": "image", "uri": "media/bdv2fail/2608504e665c71c0a60f0a25.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0173/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 173, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the cloth to the right of the toy sink", "subtask": "move the cloth to the right of the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e518077a57f79b81c7097538.png"}, {"type": "image", "uri": "media/bdv2fail/dc0da7265154d808b6b8fd75.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0174/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_06_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 174, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_06_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the pot on the right upper part of the stove.", "subtask": "put the pot on the right upper part of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/10632eb0957c40efef4f0a39.png"}, {"type": "image", "uri": "media/bdv2fail/a99bbadd50f5a1b1f94b3879.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0175/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_02_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 175, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_02_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the red fork over the right burner", "subtask": "move the red fork over the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ad5fd4b8a15678e2291a09cd.png"}, {"type": "image", "uri": "media/bdv2fail/3c1131a5d574b129f2bad18c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0176/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 176, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "take the yellow brush and put on upper left side of the table", "subtask": "take the yellow brush and put on upper left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/02a037815b1c3049a02b788e.png"}, {"type": "image", "uri": "media/bdv2fail/8bc97a917bcc57360dd525b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0177/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 177, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "unfold the cloth from the middle to top left", "subtask": "unfold the cloth from the middle to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c96a764a3bad0c31232a4b14.png"}, {"type": "image", "uri": "media/bdv2fail/f26c2e97aba3f187e95915f2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0178/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_23_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 178, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_23_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the yellow scrubber to the middle of the stove", "subtask": "move the yellow scrubber to the middle of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f36c8a64856c684950fe33dd.png"}, {"type": "image", "uri": "media/bdv2fail/46afe683a2a60d4e7e7b642e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0179/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 179, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the metal bowl between the burners.", "subtask": "move the metal bowl between the burners.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9444624b4be8dd40422809ae.png"}, {"type": "image", "uri": "media/bdv2fail/d285763935e5a334cff32fe0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0180/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 180, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the red fork between the stove and the sink", "subtask": "place the red fork between the stove and the sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a4b0da750d8bc57c7bc089e6.png"}, {"type": "image", "uri": "media/bdv2fail/42889ccf9735c917738a94e5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0181/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_70_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 181, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_70_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "remove spatula from plate sink", "subtask": "remove spatula from plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d4d70bbb1008d410fc12937a.png"}, {"type": "image", "uri": "media/bdv2fail/f10a3cb9e75f3eacd9db109f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0182/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/40", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 182, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "40", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move blue spatula to right of stove from left burner", "subtask": "move blue spatula to right of stove from left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0812e9a4727a17e46891ba5f.png"}, {"type": "image", "uri": "media/bdv2fail/7921bf8c20b4ffe262d88957.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0183/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 183, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the ladle spoon on top of the plastic silver pot", "subtask": "put the ladle spoon on top of the plastic silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e7593a242e11348a1974bc1e.png"}, {"type": "image", "uri": "media/bdv2fail/f10af344b848619c590dd50f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0184/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_47_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 184, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_47_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put cup on plate", "subtask": "put cup on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6ad1a3c7322d7f9b7d118cd.png"}, {"type": "image", "uri": "media/bdv2fail/7a0a8de7f4dbb945917e87e9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0185/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 185, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "remove the object from the orange pot", "subtask": "remove the object from the orange pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ab5913d96a641ca7d2b3f7bf.png"}, {"type": "image", "uri": "media/bdv2fail/482057f137ffb97b9914b7f7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0186/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_04_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 186, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_04_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "take the silver pot and put it underneath of the orange cloth", "subtask": "take the silver pot and put it underneath of the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/800698728a6ab05bbb50430a.png"}, {"type": "image", "uri": "media/bdv2fail/02c5858563adc22c6dddf39e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0187/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_40_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 187, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_40_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the napkin to the right of the pot and move the tomato towards it.", "subtask": "move the napkin to the right of the pot and move the tomato towards it.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c336591a6b66b41d921e61c7.png"}, {"type": "image", "uri": "media/bdv2fail/5e1cb8b87c1d92b68d150af2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0188/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_97_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 188, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_97_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move a yellow cloth to the bottom right side of the stove", "subtask": "move a yellow cloth to the bottom right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a73fbf24b4cd93e1bf761400.png"}, {"type": "image", "uri": "media/bdv2fail/dbafda0d494d99390f83bc12.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0189/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_03_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 189, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_03_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put banana around but not inside colander", "subtask": "put banana around but not inside colander", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac5998ec34d707221cff590a.png"}, {"type": "image", "uri": "media/bdv2fail/6edb1091c31cf813cc662ed3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0190/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train/32", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 190, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train", "episode_id": "32", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7dc71afe9e563f71321cd84e.png"}, {"type": "image", "uri": "media/bdv2fail/f98837c6ee86f5a7418ac017.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0191/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/51", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 191, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "51", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "put the yellow brush to the bottom right burner", "subtask": "put the yellow brush to the bottom right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e3e1dde861703d9d7156294c.png"}, {"type": "image", "uri": "media/bdv2fail/169b21375906844f27fb7ee0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0192/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_75_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 192, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_75_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the bottom right of the gray sink.", "subtask": "move the blue spoon to the bottom right of the gray sink.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/355752f3145a544ae2844058.png"}, {"type": "image", "uri": "media/bdv2fail/42e2bbe705ed69934072aa19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0193/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 193, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the corn on the lower left side of the right burner", "subtask": "put the corn on the lower left side of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3760b6926602a92f7e6e8c23.png"}, {"type": "image", "uri": "media/bdv2fail/7f4108fbe909082d4effca5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0194/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 194, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "slide the yellow spoon over to the front right corner of the stove.", "subtask": "slide the yellow spoon over to the front right corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2cc02ef2b86dd5c8959b69ac.png"}, {"type": "image", "uri": "media/bdv2fail/499bdb481bd0fa60997184ad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0195/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 195, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "remove the lid of the silver dish and put it on the upper right side of the table", "subtask": "remove the lid of the silver dish and put it on the upper right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e5e853f3b93796151fdc6af.png"}, {"type": "image", "uri": "media/bdv2fail/f7dc8eced90e464a81862ab2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0196/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_12_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 196, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_12_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 249, "role": "outcome"}}
{"state": {"task": "move the wooden box to the drawer.", "subtask": "move the wooden box to the drawer.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc288b2bbc62dbcfc71c150c.png"}, {"type": "image", "uri": "media/bdv2fail/03550d729a91a62af8648858.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0197/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_26_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 197, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_26_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "moves green spatula to left side from the middle of stove", "subtask": "moves green spatula to left side from the middle of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b45f2201d5242af289fe6654.png"}, {"type": "image", "uri": "media/bdv2fail/5bc6786c9daed88027f82a1c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0198/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 198, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the green object next to the pot to upper right side of the stove", "subtask": "move the green object next to the pot to upper right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4aba18fc079ba1309e68b5a0.png"}, {"type": "image", "uri": "media/bdv2fail/9712fc022fdc49765ac64f4d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0199/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_86_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 199, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_86_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the knife and place it behind the vessel", "subtask": "move the knife and place it behind the vessel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/be40fe33115c5c5ddadf7b33.png"}, {"type": "image", "uri": "media/bdv2fail/10c904b263c8327309b2a8e2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0200/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_01_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 200, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_01_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the blue spatula close by but not on the left burner", "subtask": "move the blue spatula close by but not on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/783b3ec5d8c46e163ee8e7c4.png"}, {"type": "image", "uri": "media/bdv2fail/b24e1d8d09e7f4325e6abbf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0201/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 201, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the silver bowl to the top left side of the table.", "subtask": "move the silver bowl to the top left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ac07bb9fabdcee6bc274b0f.png"}, {"type": "image", "uri": "media/bdv2fail/3cf39080b46015547b130b17.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0202/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 202, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c7d988fe071fb8d1d4eba40.png"}, {"type": "image", "uri": "media/bdv2fail/899b9576336f585c0eda35fa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0203/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 203, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "move the pear to the bottom right of the burner", "subtask": "move the pear to the bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/31e659ab4e8709fe59fd0106.png"}, {"type": "image", "uri": "media/bdv2fail/f291024a02fa39a39920610e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0204/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 204, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "moves the fruit to the cloth from the left of the table", "subtask": "moves the fruit to the cloth from the left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/78ed34c6cfdc5badde02afe8.png"}, {"type": "image", "uri": "media/bdv2fail/715ae8662b89c1c33922f1d0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0205/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train/42", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 205, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train", "episode_id": "42", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "take the yellow brush and put it on the lid", "subtask": "take the yellow brush and put it on the lid", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/02a037815b1c3049a02b788e.png"}, {"type": "image", "uri": "media/bdv2fail/8bc97a917bcc57360dd525b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0206/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 206, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top right to left", "subtask": "unfold the cloth from top right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4c1ef824b034bb673c751abd.png"}, {"type": "image", "uri": "media/bdv2fail/bc21fbcf6abc6f4da3d38f29.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0207/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train/29", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 207, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train", "episode_id": "29", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom left to top right", "subtask": "fold the cloth from bottom left to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a1fa89919f291a811ff6f716.png"}, {"type": "image", "uri": "media/bdv2fail/991d11d55ece97302163c24a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0208/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_07_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 208, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_07_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put blueberries close by but not on plate sink", "subtask": "put blueberries close by but not on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/59c7373c6e708915ebcd6c37.png"}, {"type": "image", "uri": "media/bdv2fail/cb2ebef068c027d40732b148.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0209/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 209, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "take the silver pot and keep it on the bottom left of the stove.", "subtask": "take the silver pot and keep it on the bottom left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dfef2f1a765cdbc617e83e2.png"}, {"type": "image", "uri": "media/bdv2fail/238b3cf466e42f28e14be0ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0210/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_67_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 210, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_67_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the right side of the silver pot from the center of the induction stove.", "subtask": "move the blue spoon to the right side of the silver pot from the center of the induction stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd4a34150bbda26b93e0d803.png"}, {"type": "image", "uri": "media/bdv2fail/3cd21797217244e8c8f8aa71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0211/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 211, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "move the toy microwave to the lower middle part of the table.", "subtask": "move the toy microwave to the lower middle part of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d31a1dd75842175c6d418392.png"}, {"type": "image", "uri": "media/bdv2fail/c13eb30d7d3bf39867b9de12.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0212/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 212, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold the cloth from left to right", "subtask": "fold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/33bb78ae4cd14d9b8a73dbd7.png"}, {"type": "image", "uri": "media/bdv2fail/703736b491f29f55b2e5347c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0213/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_29_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 213, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_29_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the silver pot from the top left corner of the stove to the stove", "subtask": "move the silver pot from the top left corner of the stove to the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/051f24ca4e6f0d672d3560ca.png"}, {"type": "image", "uri": "media/bdv2fail/aca3a0b630d2b7fd14bfef3e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0214/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 214, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "close a small box", "subtask": "close a small box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2418d9e60ad3349e6fe07d52.png"}, {"type": "image", "uri": "media/bdv2fail/73c942b2dcc8550c26354fa9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0215/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 215, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the blue object to the lower left corner of the stove", "subtask": "move the blue object to the lower left corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2a3504f68043af041b243b38.png"}, {"type": "image", "uri": "media/bdv2fail/721ca792cafa42b6fc62468f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0216/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_11_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 216, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_11_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the lower right side of the burner", "subtask": "move the blue spoon to the lower right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6f09b245e433adf0e664da97.png"}, {"type": "image", "uri": "media/bdv2fail/1e63651c2af4ee80b6b74a25.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0217/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 217, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the blue toothbrush to the right side of the burner.", "subtask": "move the blue toothbrush to the right side of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/998a936ac6f2554536887c4c.png"}, {"type": "image", "uri": "media/bdv2fail/243b022a618e6d43f2dd5efb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0218/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_03_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 218, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_03_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "place the knife on top of the orange towel", "subtask": "place the knife on top of the orange towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf3548d0665b210211f53718.png"}, {"type": "image", "uri": "media/bdv2fail/d6ea9ccaa8c7817800c47f84.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0219/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_92_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 219, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_92_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "pick up blue pen and put into drawer", "subtask": "pick up blue pen and put into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/37ead59d9ca82caff46c5012.png"}, {"type": "image", "uri": "media/bdv2fail/41250ba8586edf43d87584ff.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0220/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/41", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 220, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "41", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the yellow cloth behind the metal pot.", "subtask": "move the yellow cloth behind the metal pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fff75593c969ef48205b0ea5.png"}, {"type": "image", "uri": "media/bdv2fail/6e508445ea17a7b00fefcd1c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0221/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_08_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 221, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_08_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the spoon at the upper right of the table", "subtask": "put the spoon at the upper right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/537975833c1be12ddc9d7350.png"}, {"type": "image", "uri": "media/bdv2fail/85306a59949ce9815e0384be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0223/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_11_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 223, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_11_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "remove the cloth from the bottom right corner of the table", "subtask": "remove the cloth from the bottom right corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c9f36085fe5ebe76674c6d70.png"}, {"type": "image", "uri": "media/bdv2fail/1d71cf0ef09e96a354cf5d57.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0224/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_33_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 224, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_33_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "moves pot to upper left corner from bottom left corner.", "subtask": "moves pot to upper left corner from bottom left corner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c36533e7ee2ac3ab2fc4107.png"}, {"type": "image", "uri": "media/bdv2fail/9843fb458a76605ed1c93ff6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0225/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 225, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "moves the rubber chicken from upper left corner to bottom left corner", "subtask": "moves the rubber chicken from upper left corner to bottom left corner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9843fb458a76605ed1c93ff6.png"}, {"type": "image", "uri": "media/bdv2fail/2c36533e7ee2ac3ab2fc4107.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0226/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 226, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the silver pot to the top of the right burner", "subtask": "move the silver pot to the top of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/51aa2ba54aa20671d6b510fe.png"}, {"type": "image", "uri": "media/bdv2fail/11dcf36cec255fe24d4be854.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0227/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_00_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 227, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_00_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth to the bottom right corner of the burner", "subtask": "move the cloth to the bottom right corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3f1e7392f8ded1be8e1c0536.png"}, {"type": "image", "uri": "media/bdv2fail/b4646bb32768c7ad9c74cc42.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0228/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 228, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver spoon to the middle side of the table.", "subtask": "move the silver spoon to the middle side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b84674213e777dc0fd1805c1.png"}, {"type": "image", "uri": "media/bdv2fail/dc837253f18a1bc33bf26deb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0229/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 229, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth from the left to the bottom right of the table", "subtask": "move the cloth from the left to the bottom right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7629502823b7866901cf5b1c.png"}, {"type": "image", "uri": "media/bdv2fail/c265207fe63c9bc0271de7fc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0230/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_22_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 230, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_22_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the silver pot to the grey sink", "subtask": "move the silver pot to the grey sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/18d9ae62293640cb5a8c95b9.png"}, {"type": "image", "uri": "media/bdv2fail/aa863f51b10dddb658775fba.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0231/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_39_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 231, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_39_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "pick up blue screwdriver.", "subtask": "pick up blue screwdriver.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/93420922681c8ce81cef8b98.png"}, {"type": "image", "uri": "media/bdv2fail/cba28588f6ac3c325745016e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0232/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train/40", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 232, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train", "episode_id": "40", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the bottom side on the table", "subtask": "move the yellow cloth to the bottom side on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ccac6a90bca8eedddf50da52.png"}, {"type": "image", "uri": "media/bdv2fail/964dccb873fc3c6f314eafea.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0233/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 233, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the blue spatula out of the pan.", "subtask": "put the blue spatula out of the pan.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e6414471fd3fc9bb921d3fd.png"}, {"type": "image", "uri": "media/bdv2fail/369936818ed3f87c451d3ac2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0234/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 234, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the grey cloth to the bottom middle part of the stove.", "subtask": "move the grey cloth to the bottom middle part of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f4d612bda78736c843f1516.png"}, {"type": "image", "uri": "media/bdv2fail/67e0e7a8f1705208c81cd2de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0235/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_09_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 235, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_09_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "take the silver pot and cook it with a red pepper", "subtask": "take the silver pot and cook it with a red pepper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/77f9aeca510431f1129b3219.png"}, {"type": "image", "uri": "media/bdv2fail/6cfcb102525afcdaf5cbc00e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0236/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 236, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the blue spoon on top of the green cloth.", "subtask": "put the blue spoon on top of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/888867d29ffd0d9ab5038f7b.png"}, {"type": "image", "uri": "media/bdv2fail/63fcd9930f4272aa1004534d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0237/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 237, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot to the left burner.", "subtask": "move the silver pot to the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f2a47d69693419ed1d256563.png"}, {"type": "image", "uri": "media/bdv2fail/dba612923759fb5178cb07de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0238/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 238, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c7d988fe071fb8d1d4eba40.png"}, {"type": "image", "uri": "media/bdv2fail/7c7d988fe071fb8d1d4eba40.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0239/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 239, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "fold the cloth from left to right", "subtask": "fold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d8b2e12e74e93a6cd74ca867.png"}, {"type": "image", "uri": "media/bdv2fail/c2560ca3b2cfbfd28f1ac503.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0241/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 241, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bce3ef38009e6caccd0c94a2.png"}, {"type": "image", "uri": "media/bdv2fail/3d29fef4bfce5020e820eeb6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0242/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_13_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 242, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_13_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "put the silver pot between the stove and the sink", "subtask": "put the silver pot between the stove and the sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3234185a0bfa97411435a2a5.png"}, {"type": "image", "uri": "media/bdv2fail/000348d821b8723baadab927.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0243/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 243, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "unfold the cloth from left to right", "subtask": "unfold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/59123a67f89cc4fb302d117f.png"}, {"type": "image", "uri": "media/bdv2fail/59123a67f89cc4fb302d117f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0244/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 244, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train", "episode_id": "5", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "pick up violet allen key", "subtask": "pick up violet allen key", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/020cbd64b344ff628c55c6a1.png"}, {"type": "image", "uri": "media/bdv2fail/5810bcfb436950bfe2e244c7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0245/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_violet_Allen_key_train/27", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 245, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_violet_Allen_key_train", "episode_id": "27", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the yellow knife to the left side of the burner", "subtask": "move the yellow knife to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c1fb4f537cbe91532a67feb.png"}, {"type": "image", "uri": "media/bdv2fail/592a671a38b9e31cd6d7eed8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0246/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 246, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put yellow toy banana in blue plate", "subtask": "put yellow toy banana in blue plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eb9313a0269ac2ad9313717e.png"}, {"type": "image", "uri": "media/bdv2fail/4cd407928c4523cb7804dcd6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0247/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train/39", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 247, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train", "episode_id": "39", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put toy lemon from counter to sink", "subtask": "put toy lemon from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c85b8a4fa2b7f7ac499c9a1e.png"}, {"type": "image", "uri": "media/bdv2fail/8e871c8d4ae3e86b344448ec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0248/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/26", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 248, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "26", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "place the yellow banana to the top left corner of the plastic sink", "subtask": "place the yellow banana to the top left corner of the plastic sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70a8029ed3e32a43df6163e3.png"}, {"type": "image", "uri": "media/bdv2fail/9348e78412fd0a85d4bdebe3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0249/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 249, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "unfold the cloth from bottom left to right", "subtask": "unfold the cloth from bottom left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/842b523fd661ab43a4b026db.png"}, {"type": "image", "uri": "media/bdv2fail/ab0f6b957a6d9d462de4a686.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0250/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_11_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 250, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_11_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "moves green spatula from right side to the middle of stove", "subtask": "moves green spatula from right side to the middle of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5bc6786c9daed88027f82a1c.png"}, {"type": "image", "uri": "media/bdv2fail/b45f2201d5242af289fe6654.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0251/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 251, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "close the lip of silver pot and move to the right upper side", "subtask": "close the lip of silver pot and move to the right upper side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/21e8524b6b602ef4ee0ea89b.png"}, {"type": "image", "uri": "media/bdv2fail/b71a6c4d6d10c4a00f6c8aa8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0252/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 252, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the silver pan to the right side of the stove", "subtask": "move the silver pan to the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/060fb26aa869ed92663da79c.png"}, {"type": "image", "uri": "media/bdv2fail/19cb7c8614612be205f3df82.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0253/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_41_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 253, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_41_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue spoon and sushi to the right side of the table.", "subtask": "move the blue spoon and sushi to the right side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/98b9ab16e1120a31bab9cc59.png"}, {"type": "image", "uri": "media/bdv2fail/1b330d779384f39d4fd708a5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0254/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_01_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 254, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_01_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the orange cloth", "subtask": "move the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c2264cfb485d1fb6c8fa5980.png"}, {"type": "image", "uri": "media/bdv2fail/4d352109770ccb26952bbfb2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0255/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 255, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "remove the blue fork from the upper right corner of the burner", "subtask": "remove the blue fork from the upper right corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a44be1c6ddf79debf9727a2.png"}, {"type": "image", "uri": "media/bdv2fail/858d719c2d354107c5d56c7a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0256/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 256, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red toy fork to the right side of the pot.", "subtask": "move the red toy fork to the right side of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9aec55453901a208bc241d7.png"}, {"type": "image", "uri": "media/bdv2fail/9ac3ac418bdd7d8694d61b54.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0257/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_07_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 257, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_07_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red spoon in the middle and the left side of silver pot", "subtask": "move the red spoon in the middle and the left side of silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2449f33c97b6a9259171a677.png"}, {"type": "image", "uri": "media/bdv2fail/7baebfa4bb6e75cb389e3547.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0258/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_17_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 258, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_17_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom to top", "subtask": "fold the cloth from bottom to top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/753eb4c2fb33a3a61b49dc02.png"}, {"type": "image", "uri": "media/bdv2fail/ca38805ce248e0dc19cf00e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0259/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_12_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 259, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_12_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the silver lid on the green toy spoon.", "subtask": "put the silver lid on the green toy spoon.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/12a3982a84be10dcc2289c44.png"}, {"type": "image", "uri": "media/bdv2fail/be32103799c792ede7b439c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0260/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 260, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the upper let side of the left burner", "subtask": "remove the silver pot from the upper let side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/76ea18347bbc7a6344889203.png"}, {"type": "image", "uri": "media/bdv2fail/0e5d31ab414a9355f8035322.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0261/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_42_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 261, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_42_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the elephant to the left side of the table", "subtask": "move the elephant to the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ad068cff729074332a509944.png"}, {"type": "image", "uri": "media/bdv2fail/16043f1b7d3b61bfab9dd8c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0262/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 262, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the upper left from the lower right of the stove.", "subtask": "move the blue spoon to the upper left from the lower right of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/681f3d016aa9346d987aade2.png"}, {"type": "image", "uri": "media/bdv2fail/250740c61b1792ea2e94b3be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0263/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 263, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the silver pot to the right side of the oven top", "subtask": "move the silver pot to the right side of the oven top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/24a4902f82d292a056776e28.png"}, {"type": "image", "uri": "media/bdv2fail/6b53e5463e7e940c7d858b49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0264/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_64_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 264, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_64_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "take the silver lid off the silver pot and put it on the right side of the silver pot", "subtask": "take the silver lid off the silver pot and put it on the right side of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e07266de493f89fdd46818e.png"}, {"type": "image", "uri": "media/bdv2fail/c07720c11cb973ed6a9ed9a8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0265/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 265, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "put the blue sponge on the upper right corner of the burner.", "subtask": "put the blue sponge on the upper right corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/858d719c2d354107c5d56c7a.png"}, {"type": "image", "uri": "media/bdv2fail/0a44be1c6ddf79debf9727a2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0266/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 266, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put carrot beside but not on plate", "subtask": "put carrot beside but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79a7f3e36889ede4416f5487.png"}, {"type": "image", "uri": "media/bdv2fail/681efcafab9b969741ab6378.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0267/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_carrot_on_plate_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 267, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_carrot_on_plate_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "remove the cloth from the bottom left corner of the table", "subtask": "remove the cloth from the bottom left corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c9f36085fe5ebe76674c6d70.png"}, {"type": "image", "uri": "media/bdv2fail/1d71cf0ef09e96a354cf5d57.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0268/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_33_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 268, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_33_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot on the right side of stove", "subtask": "move the silver pot on the right side of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bfbb17c2d31319e7c2ca91ba.png"}, {"type": "image", "uri": "media/bdv2fail/113a177f72f1c467c87d34ab.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0269/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_60_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 269, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_60_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "fold the cloth from the bottom right to the left", "subtask": "fold the cloth from the bottom right to the left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2729afb4be6eb1b37d23e8ed.png"}, {"type": "image", "uri": "media/bdv2fail/ae79cc6306a586a5847e30eb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0270/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 270, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the grey plastic to the right burner.", "subtask": "move the grey plastic to the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/65d4df1fed899e90ea46c903.png"}, {"type": "image", "uri": "media/bdv2fail/ba7acda5635efff78f9133c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0271/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_33_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 271, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_33_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "pick up scissors and put into drawer", "subtask": "pick up scissors and put into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e965f9d8e06b966ff135b275.png"}, {"type": "image", "uri": "media/bdv2fail/2ffe84b552b721d6587847fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0272/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train/38", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 272, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train", "episode_id": "38", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move a pot of silver to the middle of table", "subtask": "move a pot of silver to the middle of table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/44f29983892bac03afe05a7f.png"}, {"type": "image", "uri": "media/bdv2fail/ef2a0602cd9a6cba75a2e6e9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0273/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_08_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 273, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_08_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "take the silver vessel and place it close to but not on the purple cloth.", "subtask": "take the silver vessel and place it close to but not on the purple cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/62f99b51f3c8b8557574d4a6.png"}, {"type": "image", "uri": "media/bdv2fail/e4acfcdcd6ae2ce11e376c73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0274/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_02_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 274, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_02_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the pot to the left", "subtask": "move the pot to the left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ffc9d29910b039505829b301.png"}, {"type": "image", "uri": "media/bdv2fail/76ad96f7b228ea3c988eaf84.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0275/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_40_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 275, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_40_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the blue spatula to the right side of the burner", "subtask": "move the blue spatula to the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14f4a698e896767d3f97c4b7.png"}, {"type": "image", "uri": "media/bdv2fail/7fd933337a022e5ea68b5f0c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0276/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 276, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the silver pot on the top of yellow towel in the upper right corner", "subtask": "move the silver pot on the top of yellow towel in the upper right corner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1a2dd2536892ff8e95bee79e.png"}, {"type": "image", "uri": "media/bdv2fail/c9bf8fe8df688816942e8dc9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0277/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_33_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 277, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_33_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the blue carpet to left from right of the table.", "subtask": "move the blue carpet to left from right of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2d0e2c1c80a630c58f265efd.png"}, {"type": "image", "uri": "media/bdv2fail/951bcbca10bfb008825032bd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0278/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 278, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow object from the pot to the right of the left side burner.", "subtask": "move the yellow object from the pot to the right of the left side burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0e41312cca45a06022e86177.png"}, {"type": "image", "uri": "media/bdv2fail/8eff56aaacb09f759a791e04.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0279/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 279, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the can to the left side of the stove", "subtask": "move the can to the left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc01f813c13beb6cbcd794e6.png"}, {"type": "image", "uri": "media/bdv2fail/43a134743df93288cc3e3c78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0280/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 280, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the gray bowl to the right of the burner.", "subtask": "move the gray bowl to the right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/002a2ed005f8663642682f2c.png"}, {"type": "image", "uri": "media/bdv2fail/61e44c16422e67a3b8960ece.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0281/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_60_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 281, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_60_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top right to bottom left revealing the red figure on the top of the cloth", "subtask": "unfold the cloth from top right to bottom left revealing the red figure on the top of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f2f0b8915143648795b61725.png"}, {"type": "image", "uri": "media/bdv2fail/0e34090ded0e8dc4b266fd53.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0283/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_08_train/33", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 283, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_08_train", "episode_id": "33", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "take the silver pot and put it on the top of the stove.", "subtask": "take the silver pot and put it on the top of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/838447fa7165c95c96582620.png"}, {"type": "image", "uri": "media/bdv2fail/cd18fcf7ee91946939bdeb3c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0284/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 284, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "remove the lid from the metal pot and put it underneath of the green cloth", "subtask": "remove the lid from the metal pot and put it underneath of the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/de9396e1269d42f4ff061a1c.png"}, {"type": "image", "uri": "media/bdv2fail/b7a77f5c541e9cbdd257349b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0285/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 285, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the spoon to the left from the right and put it in top of the green cloth.", "subtask": "move the spoon to the left from the right and put it in top of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf69525ae0fc9e844c647e73.png"}, {"type": "image", "uri": "media/bdv2fail/433a332fedef5d025d4d3bce.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0286/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_28_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 286, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_28_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 249, "role": "outcome"}}
{"state": {"task": "unfold the cloth from bottom to top", "subtask": "unfold the cloth from bottom to top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fda0bcce2ca7d3a894793274.png"}, {"type": "image", "uri": "media/bdv2fail/0263353f078059b3b6170f5a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0287/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_26_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 287, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_26_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "place the cloth to the right edge of the machine", "subtask": "place the cloth to the right edge of the machine", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a53b07cf775dbeaea098ee04.png"}, {"type": "image", "uri": "media/bdv2fail/c72907dcee1aeabe8735aacc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0288/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_09_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 288, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_09_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "place the red fork on the stove.", "subtask": "place the red fork on the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a4b0da750d8bc57c7bc089e6.png"}, {"type": "image", "uri": "media/bdv2fail/42889ccf9735c917738a94e5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0289/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_70_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 289, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_70_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top left to bottom right", "subtask": "unfold the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/117f4ba4cfd1d67c67686dbb.png"}, {"type": "image", "uri": "media/bdv2fail/69561471610a457138fbf739.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0290/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_04_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 290, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_04_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the green object close by but not in between the burners", "subtask": "put the green object close by but not in between the burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4585d0acc38272f46e71d7c4.png"}, {"type": "image", "uri": "media/bdv2fail/6ea37b21854ef0c98071472d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0291/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 291, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the black tape to top from bottom", "subtask": "move the black tape to top from bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fbd5256059f152bb7404ce4b.png"}, {"type": "image", "uri": "media/bdv2fail/72d30aa3d34fb9a81eedb5e2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0292/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 292, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move yellow pear from right side to the burner to the top middle .", "subtask": "move yellow pear from right side to the burner to the top middle .", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9128ec4b0e42cb861bc44e73.png"}, {"type": "image", "uri": "media/bdv2fail/490e681529805a05978a9472.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0293/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_51_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 293, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_51_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "remove the white and brown thing from the pot", "subtask": "remove the white and brown thing from the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/af0651544cc6259b9453c4bc.png"}, {"type": "image", "uri": "media/bdv2fail/6d1dd487dc20395d6a46e621.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0294/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 294, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_03_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the blue fork to the bottom right corner of the toy sink", "subtask": "move the blue fork to the bottom right corner of the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c959092d6217d1397748c032.png"}, {"type": "image", "uri": "media/bdv2fail/96784e8df9e2a8d2c2c4b2de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0295/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 295, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move pan from left to right", "subtask": "move pan from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1f434e508d902c274259396c.png"}, {"type": "image", "uri": "media/bdv2fail/4569e7869ae8d0624a3d1055.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0296/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 296, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "remove the yellow cloth from the right side of the table.", "subtask": "remove the yellow cloth from the right side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/73022d869b1beefe4cda3405.png"}, {"type": "image", "uri": "media/bdv2fail/93cc244b5dd9b6451709acd1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0297/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_28_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 297, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_28_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the green object next to but not in the silver pot", "subtask": "put the green object next to but not in the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a579474f3d1b86375cbf1b68.png"}, {"type": "image", "uri": "media/bdv2fail/f5009d123f10282ba5b8ac73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0298/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_74_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 298, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_74_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the blue fork from the upper left corner of the burner", "subtask": "remove the blue fork from the upper left corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a44be1c6ddf79debf9727a2.png"}, {"type": "image", "uri": "media/bdv2fail/858d719c2d354107c5d56c7a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0299/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 299, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the toy microwave to the bottom of the table.", "subtask": "move the toy microwave to the bottom of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a628c2c636d373ca727d8859.png"}, {"type": "image", "uri": "media/bdv2fail/ffdce5de74188e5b350235ec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0300/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 300, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "fold the cloth from the right to the left", "subtask": "fold the cloth from the right to the left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/17474648a31be063f2779a23.png"}, {"type": "image", "uri": "media/bdv2fail/17474648a31be063f2779a23.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0301/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 301, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "21", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the yellow plastic banana to the top left corner of the burner.", "subtask": "move the yellow plastic banana to the top left corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a95af49e43a45c5edb6c465e.png"}, {"type": "image", "uri": "media/bdv2fail/c051ac363eda61529c203fdf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0302/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 302, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "open low fridge", "subtask": "open low fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/821d46f9e40ba9b079289724.png"}, {"type": "image", "uri": "media/bdv2fail/e3f6a6e85b22a52744e0471c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0303/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen5_close_low_fridge_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 303, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen5_close_low_fridge_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "move the yellow cloth above the silver pot", "subtask": "move the yellow cloth above the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/989dcb783561a0d8765d1d7c.png"}, {"type": "image", "uri": "media/bdv2fail/5c906e780d7d11c71dbe7112.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0304/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_47_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 304, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_47_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the blue pot holder to the bottom left burner.", "subtask": "move the blue pot holder to the bottom left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d3c0923983210a72c9e8a47.png"}, {"type": "image", "uri": "media/bdv2fail/17f02f73338b9457674f19d4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0305/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_29_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 305, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_29_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the green toy to the left corner of the burner.", "subtask": "move the green toy to the left corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a9ba20213047b09ee58eb5fb.png"}, {"type": "image", "uri": "media/bdv2fail/7440e177733e26251e68c520.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0306/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_83_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 306, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_83_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the lid on the grey pot", "subtask": "put the lid on the grey pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/34d0128fad3c66f26c4e8fe9.png"}, {"type": "image", "uri": "media/bdv2fail/d39bcfec2d4b3e701c515546.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0307/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 307, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue spoon between the silver pot and the mushroom", "subtask": "move the blue spoon between the silver pot and the mushroom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86d87d5c3500dd85230c7cf5.png"}, {"type": "image", "uri": "media/bdv2fail/adf8de94b21e92fae9a3e85e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0308/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_95_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 308, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_95_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "fold the cloth from the bottom left to the bottom right", "subtask": "fold the cloth from the bottom left to the bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5f45467d57531e6268d0e3a6.png"}, {"type": "image", "uri": "media/bdv2fail/3e75a2d38b11dddc07b25ea7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0309/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_25_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 309, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_25_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the pot above the faucet.", "subtask": "move the pot above the faucet.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/789dd36199509b6650a94861.png"}, {"type": "image", "uri": "media/bdv2fail/6dd027b67403e580ccc0cd19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0310/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_64_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 310, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_64_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the metal bowl below the right burner.", "subtask": "move the metal bowl below the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/314bdf507e2340356d9737f0.png"}, {"type": "image", "uri": "media/bdv2fail/87f3040a964040a0fd38dc9b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0311/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_26_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 311, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_26_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "take the small blue spoon to the toy stove", "subtask": "take the small blue spoon to the toy stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e0e035ee5fd5dc53cc69789b.png"}, {"type": "image", "uri": "media/bdv2fail/b070e974963fa86b9b473135.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0312/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_110_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 312, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_110_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "picking up the pot and putting it on the yellow sponge", "subtask": "picking up the pot and putting it on the yellow sponge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eefd2a61d33f23500a58a79e.png"}, {"type": "image", "uri": "media/bdv2fail/d16b960af03243034ee49a8f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0313/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_99_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 313, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_99_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove cup from plate", "subtask": "remove cup from plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a7a4e8599971151e571b8b49.png"}, {"type": "image", "uri": "media/bdv2fail/a7a4e8599971151e571b8b49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0314/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/23", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 314, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "23", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put the silver pan on the upper right side of stove", "subtask": "put the silver pan on the upper right side of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/112623062501ef837e66e25a.png"}, {"type": "image", "uri": "media/bdv2fail/2f947d4d4c049667755a07e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0315/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 315, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the red pepper to the upper left side of the table.", "subtask": "move the red pepper to the upper left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/35e4f540ff8783bb14a295ab.png"}, {"type": "image", "uri": "media/bdv2fail/6721af84bd7556d1bed8aa59.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0316/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 316, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth to the top left of the table from the bottom left", "subtask": "move the cloth to the top left of the table from the bottom left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f92c2170a61483ff8b412b20.png"}, {"type": "image", "uri": "media/bdv2fail/3c00ab61e417dfcc122bd2fc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0317/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 317, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the yellow knife to the left side of the burner", "subtask": "move the yellow knife to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c1fb4f537cbe91532a67feb.png"}, {"type": "image", "uri": "media/bdv2fail/1c1fb4f537cbe91532a67feb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0318/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 318, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the green eggplant on the white tile backsplash", "subtask": "put the green eggplant on the white tile backsplash", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e71541678506534af88b5816.png"}, {"type": "image", "uri": "media/bdv2fail/26d4a79f77d1865e2e5cf3e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0319/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_42_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 319, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_42_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the pot below the right burner", "subtask": "move the pot below the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/789dd36199509b6650a94861.png"}, {"type": "image", "uri": "media/bdv2fail/6dd027b67403e580ccc0cd19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0320/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_64_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 320, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_64_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put an object in the silver pot", "subtask": "put an object in the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/623023751eb61284cba7b703.png"}, {"type": "image", "uri": "media/bdv2fail/0754e78ef649a155b31a1bad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0321/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 321, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "fold the cloth from the top left to the bottom right", "subtask": "fold the cloth from the top left to the bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1991bd46cbe064c2df4a8252.png"}, {"type": "image", "uri": "media/bdv2fail/8013ee1c7e3d867dcfd5644a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0322/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/30", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 322, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "30", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove the yellow knife from the right burner", "subtask": "remove the yellow knife from the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2823dcc99b327a3b4448f98d.png"}, {"type": "image", "uri": "media/bdv2fail/699326b30b4915fbdcf6f292.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0323/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_02_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 323, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_02_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the lower right of the burner", "subtask": "move the yellow cloth to the lower right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14f8b0ce3da517ba0512541a.png"}, {"type": "image", "uri": "media/bdv2fail/59dd0283730462fdbfcd5218.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0324/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_18_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 324, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_18_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the green toy to the top edge of the cloth from the left of the table.", "subtask": "move the green toy to the top edge of the cloth from the left of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e8f418c4d9bdf0fc7f0a5305.png"}, {"type": "image", "uri": "media/bdv2fail/399c74ecbdf93fd99c244ccb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0325/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 325, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "24", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "move bowl to lower left of cooktop", "subtask": "move bowl to lower left of cooktop", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f11fc6b1e3e81ccfe9e69a2c.png"}, {"type": "image", "uri": "media/bdv2fail/752375ba37be8ca2533a9202.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0326/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_93_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 326, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_93_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ab1db45c3fa1f2de80aa757f.png"}, {"type": "image", "uri": "media/bdv2fail/e1eedd02af5c6c03881192b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0327/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/49", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 327, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "49", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the yellow cloth and place it in the lower right corner of the table", "subtask": "remove the silver pot from the yellow cloth and place it in the lower right corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/26b250a85408ec994349a45a.png"}, {"type": "image", "uri": "media/bdv2fail/a86bfea41967df9f35b88d61.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0328/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 328, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "take the fried chicken and put it below the right burner", "subtask": "take the fried chicken and put it below the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e4c90305b61075a4f27bf2b0.png"}, {"type": "image", "uri": "media/bdv2fail/f877956c7d60302655699eb4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0329/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 329, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the blue spoon to the center of the burner.", "subtask": "put the blue spoon to the center of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a11e82fa3b7b4b4505aa0993.png"}, {"type": "image", "uri": "media/bdv2fail/d2ba3f4d0d1d4c85b9914e41.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0330/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 330, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the yellow object to the lower right side of the desk", "subtask": "move the yellow object to the lower right side of the desk", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/00382afac86ab19d782ef023.png"}, {"type": "image", "uri": "media/bdv2fail/00382afac86ab19d782ef023.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0331/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 331, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/57578533979e054da972be29.png"}, {"type": "image", "uri": "media/bdv2fail/ed028fcec376804e6ce64226.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0332/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 332, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "24", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the yellow knife to the bottom middle part of the stove", "subtask": "move the yellow knife to the bottom middle part of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ff76cb4da0d554edd36bd7c8.png"}, {"type": "image", "uri": "media/bdv2fail/bf53c4f8f687a9663acab521.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0333/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 333, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put spatula on plate sink", "subtask": "put spatula on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f10a3cb9e75f3eacd9db109f.png"}, {"type": "image", "uri": "media/bdv2fail/d4d70bbb1008d410fc12937a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0334/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/40", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 334, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "40", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the orange fish to the bottom of the stove", "subtask": "move the orange fish to the bottom of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1528ce59796d64bf2f45af02.png"}, {"type": "image", "uri": "media/bdv2fail/f5714f57d0a8315a3109e337.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0335/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 335, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to left", "subtask": "fold the cloth from bottom right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6af4aadf0c0e8245828ec0a1.png"}, {"type": "image", "uri": "media/bdv2fail/896a43e4254892aa07505650.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0336/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_22_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 336, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_22_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the orange towel from the top left to the centre of the table.", "subtask": "move the orange towel from the top left to the centre of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/48eab26340f105a65c972f11.png"}, {"type": "image", "uri": "media/bdv2fail/b919347f16bf703468eb7c40.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0337/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_17_train/22", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 337, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_17_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "take the orange spoon to the left side of the burner", "subtask": "take the orange spoon to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a4f5aa85aa9812d84d9c123f.png"}, {"type": "image", "uri": "media/bdv2fail/956140b49fb2a3b519957bbe.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0338/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_17_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 338, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_17_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "place the spatula on the green towel", "subtask": "place the spatula on the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/370c4f352ab6342c3e79e5c4.png"}, {"type": "image", "uri": "media/bdv2fail/9ca5b496862b47e563e6a034.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0339/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 339, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move blue spoon to the bottom left corner of the burners", "subtask": "move blue spoon to the bottom left corner of the burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/73cf353cd0a0d5851fdd2963.png"}, {"type": "image", "uri": "media/bdv2fail/51928347ebfc3b06eab13676.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0340/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 340, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "place the blue plastic toy on the front left edge of the stove", "subtask": "place the blue plastic toy on the front left edge of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a49ff3162c5c99ba207f9477.png"}, {"type": "image", "uri": "media/bdv2fail/9158e2ea2fd8b8337b56cea7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0341/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_30_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 341, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_30_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the bottom corner of the side of the pot.", "subtask": "move the blue spoon to the bottom corner of the side of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ae91a1619af58fbb83ab210c.png"}, {"type": "image", "uri": "media/bdv2fail/f1e2447f5e1c368f1584c44c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0342/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 342, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move blue napkin to the upper right of table", "subtask": "move blue napkin to the upper right of table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/21a437a7f051a799352f63fa.png"}, {"type": "image", "uri": "media/bdv2fail/8c128cc0e03be33fcdb6d9aa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0343/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 343, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the blue cup into the yellow basket", "subtask": "put the blue cup into the yellow basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b321bf8039405b2c9c4ac518.png"}, {"type": "image", "uri": "media/bdv2fail/fb54d56184efbb83e25a2d23.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0344/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/67", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 344, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "67", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move blue doll from top of the drawer and place it close by but not on the table", "subtask": "move blue doll from top of the drawer and place it close by but not on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ad6671a2052a36849b18cad.png"}, {"type": "image", "uri": "media/bdv2fail/65e4ea24f5339e34c0d3136c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0345/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 345, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 249, "role": "outcome"}}
{"state": {"task": "take the lid and put it on the stove", "subtask": "take the lid and put it on the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8d38b21457bab8bf300762b.png"}, {"type": "image", "uri": "media/bdv2fail/2541d96eb31e2960043b6478.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0346/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 346, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the cloth to the middle from the top edge of the table.", "subtask": "move the cloth to the middle from the top edge of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c52c5f804f16982ef87d98b9.png"}, {"type": "image", "uri": "media/bdv2fail/f98fa713ba1b7dfdeb87df89.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0347/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 347, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom left corner of the burners.", "subtask": "move the silver pot to the bottom left corner of the burners.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/73cf353cd0a0d5851fdd2963.png"}, {"type": "image", "uri": "media/bdv2fail/51928347ebfc3b06eab13676.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0349/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 349, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the spatula in the upper left corner of the table.", "subtask": "move the spatula in the upper left corner of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e49513dc0dc844d942fd119f.png"}, {"type": "image", "uri": "media/bdv2fail/91f8105116360b7632a8ee77.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0350/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_02_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 350, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_02_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top left to left", "subtask": "unfold the cloth from top left to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f1bfee9fd3421583484942e5.png"}, {"type": "image", "uri": "media/bdv2fail/f1bfee9fd3421583484942e5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0351/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_07_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 351, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_07_train", "episode_id": "11", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the silver pot on the yellow cloth", "subtask": "place the silver pot on the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/559aa7d101f4a7f1ac319d9b.png"}, {"type": "image", "uri": "media/bdv2fail/c44d0b2e98783b9fbbec9ee1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0352/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 352, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the blue cloth to the upper left corner of the stove.", "subtask": "move the blue cloth to the upper left corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c480b856b155bdef3d96843.png"}, {"type": "image", "uri": "media/bdv2fail/24913f5b7db5a48dce285510.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0353/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_93_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 353, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_93_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove spatula from cutting board", "subtask": "remove spatula from cutting board", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/51ac62691acd8fce98355fc9.png"}, {"type": "image", "uri": "media/bdv2fail/51ac62691acd8fce98355fc9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0354/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_spatula_on_cutting_board_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 354, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_spatula_on_cutting_board_train", "episode_id": "2", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the cloth to the right side of the stove", "subtask": "move the cloth to the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/480477f84ed49cbd7e732027.png"}, {"type": "image", "uri": "media/bdv2fail/2abf6236324b4b865d01ed5a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0355/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_14_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 355, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_14_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "place the spoon at the right side of the stove.", "subtask": "place the spoon at the right side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8d3824123baf606e3bbcc846.png"}, {"type": "image", "uri": "media/bdv2fail/0f5154b8f3c8ba368e0397e4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0356/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_35_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 356, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_35_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the knife on the bottom right corner of the burner", "subtask": "move the knife on the bottom right corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b12600085f1b3fedb1f06c91.png"}, {"type": "image", "uri": "media/bdv2fail/c6b02a855e0038a796ef6352.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0357/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_78_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 357, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_78_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the bottom right of the burner", "subtask": "move the yellow cloth to the bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/112df704260b5e2cb3057c14.png"}, {"type": "image", "uri": "media/bdv2fail/eb200e445e07c6be11b99dd4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0358/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 358, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9d4f3f805e9e2e9d6e60adcc.png"}, {"type": "image", "uri": "media/bdv2fail/a46a657d90cfb88bc5dbc5ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0359/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 359, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "fold the cloth from left to right", "subtask": "fold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2fb103d8c334402d33d2ec7e.png"}, {"type": "image", "uri": "media/bdv2fail/7aea53b84145187e352b0936.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0360/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_27_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 360, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_27_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the plastic toy egg to the bottom left of the stove.", "subtask": "move the plastic toy egg to the bottom left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/983f41bd20a6a51e416e8cd0.png"}, {"type": "image", "uri": "media/bdv2fail/05d7c516999e114f886ad3fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0361/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 361, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the top of the ice cream cone", "subtask": "move the yellow cloth to the top of the ice cream cone", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6587edd0e533db315beeedfb.png"}, {"type": "image", "uri": "media/bdv2fail/0043486c552bff5e13c89b03.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0363/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 363, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the green monkey toy to the middle bottom of the drawer", "subtask": "move the green monkey toy to the middle bottom of the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c229c1ccf96ee54df909ea64.png"}, {"type": "image", "uri": "media/bdv2fail/2b71a0a9260801b6748975d3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0364/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 364, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold the cloth from the right to the left", "subtask": "fold the cloth from the right to the left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/17474648a31be063f2779a23.png"}, {"type": "image", "uri": "media/bdv2fail/7bfd59d26faadf03dd8e9526.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0365/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 365, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the cloth from right to left", "subtask": "move the cloth from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/924374499e8937a81bdd5a4a.png"}, {"type": "image", "uri": "media/bdv2fail/74681f140d9c5ea2361a859f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0366/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 366, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "take the toy sink and place it on top of the yellow cloth.", "subtask": "take the toy sink and place it on top of the yellow cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d9fa9b580af0720c76fe6ae0.png"}, {"type": "image", "uri": "media/bdv2fail/01f295430c8289188483bb1e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0367/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 367, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "put the blue spoon above the violet cloth", "subtask": "put the blue spoon above the violet cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/abf5c33232ed65bd294b7dfb.png"}, {"type": "image", "uri": "media/bdv2fail/4ff4a6295eef675332a55bb0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0368/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 368, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the plastic pink cup on the yellow dish drying rack.", "subtask": "put the plastic pink cup on the yellow dish drying rack.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c583f095c9071d4e3bca7aa.png"}, {"type": "image", "uri": "media/bdv2fail/ed8be19109dc9a2b044da7aa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0369/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 369, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove the lid from the silver bowl", "subtask": "remove the lid from the silver bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ee7766f25a37ddd4e16518e.png"}, {"type": "image", "uri": "media/bdv2fail/3b7f1b960425f354fd7f1f0a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0370/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 370, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the yellow object below the right burner and close to spoon.", "subtask": "move the yellow object below the right burner and close to spoon.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5406d3fac0e9ca218c4d4098.png"}, {"type": "image", "uri": "media/bdv2fail/57d35c8e862952f255dfb1e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0371/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_48_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 371, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_48_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "unfold the cloth from bottom right to top left", "subtask": "unfold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1a243495b1b94ea00ea1c974.png"}, {"type": "image", "uri": "media/bdv2fail/1aeef69b1bb3fe36258b1dd0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0372/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 372, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the green cloth below the silver pan on the bottom left of burner", "subtask": "move the green cloth below the silver pan on the bottom left of burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/487b70f81eccf9f869339f1e.png"}, {"type": "image", "uri": "media/bdv2fail/c47647d7a728c6b4c6dc554b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0373/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 373, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the blue dishrag to the right side of the burner", "subtask": "move the blue dishrag to the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9a6418fb9aada573bf773f2.png"}, {"type": "image", "uri": "media/bdv2fail/bfd5bcd54254f2b401a88dbf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0374/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 374, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the silver pot lid from the top of the silver pot", "subtask": "remove the silver pot lid from the top of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e1362c5991ab6987d106102f.png"}, {"type": "image", "uri": "media/bdv2fail/58e7d8debd78055fa8271394.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0375/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 375, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the violet cloth in the bottom right corner of the toy sink", "subtask": "put the violet cloth in the bottom right corner of the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ecf2cf92e22bf08d966291b0.png"}, {"type": "image", "uri": "media/bdv2fail/1463de8f48b1d9c93229d0de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0376/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_84_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 376, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_84_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the green bell pepper to the left burner beside the spatula", "subtask": "move the green bell pepper to the left burner beside the spatula", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2b6eb3fb7274f79a840a11f4.png"}, {"type": "image", "uri": "media/bdv2fail/0867b5d8b5fc06d1ef468a0c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0377/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 377, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the orange toy to the left side of the table from the middle upper part", "subtask": "move the orange toy to the left side of the table from the middle upper part", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}, {"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0378/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 378, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upperleft of table", "subtask": "move the silver pot to the upperleft of table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/411c69c79a0e1233cc485d4c.png"}, {"type": "image", "uri": "media/bdv2fail/411c69c79a0e1233cc485d4c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0379/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 379, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the red bell pepper into bottom right of the burner", "subtask": "move the red bell pepper into bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c829780dbb0b036c7ea0e29.png"}, {"type": "image", "uri": "media/bdv2fail/4270220db5bbf7711277da5e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0380/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_78_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 380, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_78_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d1da4e3e82a64df97e629245.png"}, {"type": "image", "uri": "media/bdv2fail/686a35b44e6332527305ead3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0381/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_03_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 381, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_03_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "remove the green vegetable from the right side of the burner", "subtask": "remove the green vegetable from the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec5cb938ff4a1a745a46f539.png"}, {"type": "image", "uri": "media/bdv2fail/ec5cb938ff4a1a745a46f539.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0382/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 382, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train", "episode_id": "11", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove the chocolate bar from top of the orange cloth", "subtask": "remove the chocolate bar from top of the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eab6540223a18136a2aac038.png"}, {"type": "image", "uri": "media/bdv2fail/70935790f9eaa30560a6a940.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0383/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 383, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "pick up blue pen and put into drawer", "subtask": "pick up blue pen and put into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0864fcb14d1d4ef0fce56ddc.png"}, {"type": "image", "uri": "media/bdv2fail/4c9b306394f55a58dfec647c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0384/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/28", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 384, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "28", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "pick up red srewdriver", "subtask": "pick up red srewdriver", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/93420922681c8ce81cef8b98.png"}, {"type": "image", "uri": "media/bdv2fail/cba28588f6ac3c325745016e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0385/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train/40", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 385, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train", "episode_id": "40", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "put the red fork close by but not on the green cloth", "subtask": "put the red fork close by but not on the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/42bd23e905406794665e8927.png"}, {"type": "image", "uri": "media/bdv2fail/70837f42c84534f5ea69ad69.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0386/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 386, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "cover the silver pot and put the lid on the left side of the table", "subtask": "cover the silver pot and put the lid on the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9288175aa9efc795575e0052.png"}, {"type": "image", "uri": "media/bdv2fail/f814cf96c5dcf1a5956d47b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0387/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 387, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the blue towel from the left burner to the right burner.", "subtask": "move the blue towel from the left burner to the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0475b18118c518a6e66c64b7.png"}, {"type": "image", "uri": "media/bdv2fail/e9c678dba3c0355a6ece7fa3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0388/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 388, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the red fork on top of the blue cloth", "subtask": "put the red fork on top of the blue cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79bbdba096a58685f6e196c3.png"}, {"type": "image", "uri": "media/bdv2fail/cc04978cf358f532cd05cd53.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0390/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_43_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 390, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_43_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upper right side of the table", "subtask": "move the silver pot to the upper right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/de5de53176cd3a6a25e8c9e5.png"}, {"type": "image", "uri": "media/bdv2fail/a230e0f3533bec9f651c9fc7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0391/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 391, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the bread to the top of the left burner", "subtask": "move the bread to the top of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d8fe49c3e772f22e6f6be79c.png"}, {"type": "image", "uri": "media/bdv2fail/d45104f99887632d068fbdbb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0392/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 392, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the corn to the bottom left of the burner", "subtask": "move the corn to the bottom left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d98f4911165ac3058c6d1bdd.png"}, {"type": "image", "uri": "media/bdv2fail/5b3b60ea9028615dee1445be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0393/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 393, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the orange toy from the right side of the table to the middle upper part", "subtask": "move the orange toy from the right side of the table to the middle upper part", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/92375aa847b78ec7040b05b6.png"}, {"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0394/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 394, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "remove cup from plate", "subtask": "remove cup from plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6b9f1ba9d9e66b13306a4f11.png"}, {"type": "image", "uri": "media/bdv2fail/6b9f1ba9d9e66b13306a4f11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0395/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/27", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 395, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "27", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the green fork to left of the silver pot", "subtask": "move the green fork to left of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4444af3ac5316c8102f15350.png"}, {"type": "image", "uri": "media/bdv2fail/ddd74a1844da9611a7fbf8b5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0396/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 396, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the silver pot between the stove and the blue sponge", "subtask": "put the silver pot between the stove and the blue sponge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1290672c526271caffc13fcd.png"}, {"type": "image", "uri": "media/bdv2fail/dbd9643f67868490ac611e53.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0397/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_15_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 397, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_15_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the small silver pot to the lower left corner of the stove.", "subtask": "move the small silver pot to the lower left corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/344616d784774ec413d342e1.png"}, {"type": "image", "uri": "media/bdv2fail/50035b9dd3d241c717dad202.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0398/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 398, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "0", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the blue brush to the upper right of the green sponge.", "subtask": "move the blue brush to the upper right of the green sponge.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/24887e88fd1ff5baede16f64.png"}, {"type": "image", "uri": "media/bdv2fail/e1172e7d277e9ba13b760c50.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0399/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_61_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 399, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_61_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red fork to the left side of the burner", "subtask": "move the red fork to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/69e7c3d68d65043fd2ebfa4a.png"}, {"type": "image", "uri": "media/bdv2fail/baa8a599132a52b38d993198.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0400/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 400, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom to top", "subtask": "fold the cloth from bottom to top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/34b31956db1ce63646f11861.png"}, {"type": "image", "uri": "media/bdv2fail/21fc4f3f8ee77da70a781ff2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0401/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 401, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue brush to the top of the table", "subtask": "move the blue brush to the top of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e990b26ad12326329718e951.png"}, {"type": "image", "uri": "media/bdv2fail/24c2170dff439d67a0cfccbd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0402/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 402, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "moves the silver lid to the left side of a green cloth from the upper side of the table, tin", "subtask": "moves the silver lid to the left side of a green cloth from the upper side of the table, tin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d6dfa1d37a2b0d7c2c0d59b.png"}, {"type": "image", "uri": "media/bdv2fail/572fff386ad7b684187377a9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0403/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 403, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 255, "role": "outcome"}}
{"state": {"task": "move the toy cheese into the toy sink", "subtask": "move the toy cheese into the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dda3c89983bad438cea10f4d.png"}, {"type": "image", "uri": "media/bdv2fail/413372a83226654b018cea8e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0404/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 404, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6c28b89efe20cd79957ea34.png"}, {"type": "image", "uri": "media/bdv2fail/bf7cdaf9abe2e267c08b44eb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0405/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_13_train/38", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 405, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_13_train", "episode_id": "38", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "put beet in pot sink", "subtask": "put beet in pot sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/28b6fc842d49b0287006cc9f.png"}, {"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0406/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 406, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put the can of soup at the bottom of the left burner", "subtask": "put the can of soup at the bottom of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3ade824fdf797412725045a9.png"}, {"type": "image", "uri": "media/bdv2fail/3d625433c1ca074e05da8fee.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0407/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 407, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold the cloth from top right to bottom left", "subtask": "fold the cloth from top right to bottom left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/396c9fdbd8ee544563b74ec6.png"}, {"type": "image", "uri": "media/bdv2fail/9b482be9a9ea2166fc7eb6f0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0408/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_14_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 408, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_14_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "place the yellow banana to the top left corner of the stovetop.", "subtask": "place the yellow banana to the top left corner of the stovetop.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70a8029ed3e32a43df6163e3.png"}, {"type": "image", "uri": "media/bdv2fail/9348e78412fd0a85d4bdebe3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0409/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 409, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "take the pot and put it on the green towel", "subtask": "take the pot and put it on the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d3163e550be9737fab30421.png"}, {"type": "image", "uri": "media/bdv2fail/ae181b53b43ecae1217cc5e0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0410/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 410, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the cloth to right from left", "subtask": "move the cloth to right from left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e3947a77050cb45c5876be82.png"}, {"type": "image", "uri": "media/bdv2fail/4798d42389e06450c4a519dd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0411/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 411, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the yellow duck on the drawer", "subtask": "put the yellow duck on the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d4f4aad820e54238a4d4bf87.png"}, {"type": "image", "uri": "media/bdv2fail/cadf3573d63e2515c5e0acd8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0412/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/30", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 412, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "30", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the lemon on the left side of the stove.", "subtask": "move the lemon on the left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6298e643302c41acc33c2f7d.png"}, {"type": "image", "uri": "media/bdv2fail/fc337b352444ee9b6bf7ffd8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0413/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_24_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 413, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_24_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the blue brush to the top left of the stove", "subtask": "move the blue brush to the top left of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/47febdcd44b00f869f4a49dc.png"}, {"type": "image", "uri": "media/bdv2fail/b8af5ecdad8ad1771ef5311a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0414/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 414, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the silver pot on the orange cloth", "subtask": "put the silver pot on the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4162fa9a633e3827ad2814cc.png"}, {"type": "image", "uri": "media/bdv2fail/99e3395d6c2f521032805698.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0415/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 415, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "moves the ribbon to right of the cloth from top.", "subtask": "moves the ribbon to right of the cloth from top.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4ebbc64e9b733f0257e30a7e.png"}, {"type": "image", "uri": "media/bdv2fail/43eea7b49ab6e0a76025341e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0416/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/34", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 416, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "34", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the red pepper around but not inside the pot", "subtask": "put the red pepper around but not inside the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/58674116eedb2746df95954d.png"}, {"type": "image", "uri": "media/bdv2fail/eed54a2f825d6c9b9913a180.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0417/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_90_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 417, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_90_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the pot close to but not on the napkin.", "subtask": "put the pot close to but not on the napkin.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2d84cf6c2cdd10010e8845ca.png"}, {"type": "image", "uri": "media/bdv2fail/d55d5b84b83ab3e4a0c2ddc7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0418/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_01_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 418, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_01_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the yellow cloth to the left side of the silver pot", "subtask": "put the yellow cloth to the left side of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4812e3a3de9fb9f63825b425.png"}, {"type": "image", "uri": "media/bdv2fail/d25b1ee6725505706d97dbda.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0419/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 419, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "take the silver pot to the left side of the left burner", "subtask": "take the silver pot to the left side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/84a7a0fa0c9d8193ab4633a9.png"}, {"type": "image", "uri": "media/bdv2fail/bcad9b238f6ede573da8bc0a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0420/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 420, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the yellow spoon to the bottom part between the two burners.", "subtask": "move the yellow spoon to the bottom part between the two burners.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c1298032a108bbb4f814c176.png"}, {"type": "image", "uri": "media/bdv2fail/714b479e2a29064bab0aa51f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0421/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_108_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 421, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_108_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "take the banana and put it on the right side of the table", "subtask": "take the banana and put it on the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/57fae9750910a27467472768.png"}, {"type": "image", "uri": "media/bdv2fail/cc562e0ad1c2faf55838acc6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0422/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 422, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train", "episode_id": "18", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "turn toy banana front to right.", "subtask": "turn toy banana front to right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6fb5a93e0128640e0b986cc1.png"}, {"type": "image", "uri": "media/bdv2fail/ee00a5a7595de6be9cd6df22.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0423/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 423, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train", "episode_id": "19", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the left side on the table", "subtask": "move the yellow cloth to the left side on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4ed930424456f069ea1a9cec.png"}, {"type": "image", "uri": "media/bdv2fail/e3b5431d693e48fb3df9e160.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0424/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_25_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 424, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_25_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove blueberries from plate sink", "subtask": "remove blueberries from plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cb2ebef068c027d40732b148.png"}, {"type": "image", "uri": "media/bdv2fail/59c7373c6e708915ebcd6c37.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0425/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 425, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the orange cloth from the left to the grey sink", "subtask": "move the orange cloth from the left to the grey sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/074338752e9d350bb97a6c61.png"}, {"type": "image", "uri": "media/bdv2fail/3337c2945481b99ef0c3388d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0426/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_25_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 426, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_25_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the green object from the right side of the burner into the silver pot", "subtask": "move the green object from the right side of the burner into the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5d10f6284b38b7238966ff9.png"}, {"type": "image", "uri": "media/bdv2fail/b7af6187125aa804960e2271.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0427/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 427, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "put the silver pot on the red and yellow knobs nearby", "subtask": "put the silver pot on the red and yellow knobs nearby", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14c5a0d430c195739fc480ee.png"}, {"type": "image", "uri": "media/bdv2fail/e9b3e7d0d0e1360d48d4bbc3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0428/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_06_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 428, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_06_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the orange towel to the right bottom corner of the burner", "subtask": "move the orange towel to the right bottom corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f85f8e4d2a4047e24ca227b9.png"}, {"type": "image", "uri": "media/bdv2fail/2f228138c0dda3ce60ec737c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0429/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 429, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom left corner of the table", "subtask": "move the silver pot to the bottom left corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a5186aad93ff22970a93fdfb.png"}, {"type": "image", "uri": "media/bdv2fail/a4f42c4533299b7daea74964.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0430/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 430, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put cup on plate", "subtask": "put cup on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d70649bd691d1df11cb9724.png"}, {"type": "image", "uri": "media/bdv2fail/9c040fcc7566f5e61c9c6610.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0431/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/33", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 431, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "33", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "pick up the red toolbox", "subtask": "pick up the red toolbox", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/020cbd64b344ff628c55c6a1.png"}, {"type": "image", "uri": "media/bdv2fail/5810bcfb436950bfe2e244c7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0432/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_violet_Allen_key_train/27", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 432, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_violet_Allen_key_train", "episode_id": "27", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the spoon and put it between the burners", "subtask": "move the spoon and put it between the burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/624fcc2bfce1768bb84ae25f.png"}, {"type": "image", "uri": "media/bdv2fail/28312b4a363f7ffc31924b54.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0433/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 433, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the red knobs to the bottom right of the burner.", "subtask": "move the red knobs to the bottom right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/33970f2384e0ff09de0b4e85.png"}, {"type": "image", "uri": "media/bdv2fail/721d977520e27aa1272bf3b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0434/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_11_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 434, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_11_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow knife to the top of the left burner", "subtask": "move the yellow knife to the top of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c835af7b3f50870e2880974.png"}, {"type": "image", "uri": "media/bdv2fail/de985a24f417c71fc07959d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0435/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 435, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver bowl onto the yellow cloth on the top left of the burner", "subtask": "move the silver bowl onto the yellow cloth on the top left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f575b1de5280050931f5bc1f.png"}, {"type": "image", "uri": "media/bdv2fail/d1c3c9eb99b7e57a4ad8ac2b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0436/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_28_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 436, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_28_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "place tomato inside towel", "subtask": "place tomato inside towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4811e7b50ac3b7897405ed16.png"}, {"type": "image", "uri": "media/bdv2fail/a2f85553fc00378500d441fc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0437/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_97_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 437, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_97_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the sponge above the left burner.", "subtask": "move the sponge above the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/75ac8805e4280241179870cd.png"}, {"type": "image", "uri": "media/bdv2fail/c644e8e7429bf5ca2c5c9cfd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0438/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_75_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 438, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_75_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue plastic item to the corner near the vessel.", "subtask": "move the blue plastic item to the corner near the vessel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91b29402a11de413e7d29c07.png"}, {"type": "image", "uri": "media/bdv2fail/56f181de99b94497aab18c27.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0439/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_03_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 439, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_03_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow bell pepper in front of the drawer", "subtask": "move the yellow bell pepper in front of the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9453cabac20bab1bc926b8dd.png"}, {"type": "image", "uri": "media/bdv2fail/95219641a8a28212a302ef30.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0440/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/34", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 440, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "34", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "pick up the banana and put it between the burners", "subtask": "pick up the banana and put it between the burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/125b5342fba4fb88cb682373.png"}, {"type": "image", "uri": "media/bdv2fail/de6783c3987d10e42ff9df7e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0441/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 441, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom of the green cloth.", "subtask": "move the silver pot to the bottom of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8339c0e465a9ce5e768cde06.png"}, {"type": "image", "uri": "media/bdv2fail/058e5b0114c6c0797faa1b07.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0442/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 442, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth to the left of the metal tool.", "subtask": "move the cloth to the left of the metal tool.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/20030bbeac881a51e0322251.png"}, {"type": "image", "uri": "media/bdv2fail/54d4853d62d68aca90950647.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0443/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_04_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 443, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_04_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the blue fork underneath of the orange cloth", "subtask": "move the blue fork underneath of the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4cab7430981ffa604610e270.png"}, {"type": "image", "uri": "media/bdv2fail/316fdb7ecf16c7a7ebd0847d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0444/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_97_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 444, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_97_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put cucumber in cup", "subtask": "put cucumber in cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d6929ad26a1e6f3074468f54.png"}, {"type": "image", "uri": "media/bdv2fail/fe9a55faeb734f6a89b59702.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0445/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/30", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 445, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "30", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the wood brush to the bottom of the right burner", "subtask": "move the wood brush to the bottom of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e7aa2d3d6c04a7697569855a.png"}, {"type": "image", "uri": "media/bdv2fail/7744aa9185bf4bb51bdccb68.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0446/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_33_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 446, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_33_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top to bottom", "subtask": "unfold the cloth from top to bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e8d74c530d63a73e98553491.png"}, {"type": "image", "uri": "media/bdv2fail/e8d74c530d63a73e98553491.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0447/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_03_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 447, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_03_train", "episode_id": "1", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the cloth to the top left of the table", "subtask": "move the cloth to the top left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/89f88f47e3568f745464b541.png"}, {"type": "image", "uri": "media/bdv2fail/512d4104ad87ec16d1eff9d4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0448/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 448, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "remove cup from plate", "subtask": "remove cup from plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7a0a8de7f4dbb945917e87e9.png"}, {"type": "image", "uri": "media/bdv2fail/a6ad1a3c7322d7f9b7d118cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0449/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 449, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "place the squash inside the stove", "subtask": "place the squash inside the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4c13c10f869e711c48357af2.png"}, {"type": "image", "uri": "media/bdv2fail/0bebd2a16b3c27a3f429f688.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0450/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_33_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 450, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_33_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the ladle around but not on the upper side of the table", "subtask": "move the ladle around but not on the upper side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/37cf229759e286a12b595942.png"}, {"type": "image", "uri": "media/bdv2fail/95aa020ef71ae1fb299009d3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0451/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_06_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 451, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_06_train", "episode_id": "20", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "put the yellow scrubber near but not on the green cloth.", "subtask": "put the yellow scrubber near but not on the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/def3e801c49976c0d1f75979.png"}, {"type": "image", "uri": "media/bdv2fail/1c99f5cb2750d538b55f158b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0452/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 452, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the blue cloth to the upper left of the table", "subtask": "move the blue cloth to the upper left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9cefbfdf0e942418f2953e94.png"}, {"type": "image", "uri": "media/bdv2fail/f51c5244623bca86fe29bd22.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0453/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 453, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "19", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "fold the cloth from top right top to center of the cloth", "subtask": "fold the cloth from top right top to center of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ce0c9bccd3f3409777460a4.png"}, {"type": "image", "uri": "media/bdv2fail/9ce0c9bccd3f3409777460a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0454/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_23_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 454, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_23_train", "episode_id": "13", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "pick up blue pen and put into drawer", "subtask": "pick up blue pen and put into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/554ff0f57cc96cf8ecaac79c.png"}, {"type": "image", "uri": "media/bdv2fail/939a1325285b0f65da8d1e9c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0455/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/38", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 455, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "38", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "moves spatula the spatula from right side to the middle of the stove.", "subtask": "moves spatula the spatula from right side to the middle of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5bc6786c9daed88027f82a1c.png"}, {"type": "image", "uri": "media/bdv2fail/b45f2201d5242af289fe6654.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0456/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 456, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the orange fork on top of the green cloth", "subtask": "move the orange fork on top of the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7089ec6ffd01b8ca61834720.png"}, {"type": "image", "uri": "media/bdv2fail/f7b5747946910c69979765a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0457/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 457, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the steel bowl on the left side above the right burner.", "subtask": "move the steel bowl on the left side above the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ddd85083e2e2dfeb9a04114e.png"}, {"type": "image", "uri": "media/bdv2fail/265a75f3a2f747e74f9bf53f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0458/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_41_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 458, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_41_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom to top", "subtask": "fold the cloth from bottom to top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/81e13044a017f8b20744ae28.png"}, {"type": "image", "uri": "media/bdv2fail/4c097506b98512bd4534899c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0459/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_21_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 459, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_21_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6a8fe056cc5a0cd4a151e15.png"}, {"type": "image", "uri": "media/bdv2fail/b6a8fe056cc5a0cd4a151e15.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0460/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/57", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 460, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "57", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "pick the yellow and white knife and put it on the left side of the hob", "subtask": "pick the yellow and white knife and put it on the left side of the hob", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bbb6dcd64df2e14cadebc265.png"}, {"type": "image", "uri": "media/bdv2fail/7e14d152bc20fb7e8286c955.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0461/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 461, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "put spoon in bowl sink", "subtask": "put spoon in bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc645d1faf0738f842573afc.png"}, {"type": "image", "uri": "media/bdv2fail/d2868c670a67fb95bee280e0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0462/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 462, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "24", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the cheese to the right of the yellow toy", "subtask": "move the cheese to the right of the yellow toy", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8ecd81906d7736b2c835391.png"}, {"type": "image", "uri": "media/bdv2fail/9f5941add05dd616c3c28968.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0463/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_91_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 463, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_91_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "placing the purple towel on the left side of the burner", "subtask": "placing the purple towel on the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/662024e66a5d8571b8fd769c.png"}, {"type": "image", "uri": "media/bdv2fail/5918c40580ab49cc27dfc6a3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0464/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 464, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the green toy from the top edge of the cloth to the left of the table.", "subtask": "move the green toy from the top edge of the cloth to the left of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/399c74ecbdf93fd99c244ccb.png"}, {"type": "image", "uri": "media/bdv2fail/e8f418c4d9bdf0fc7f0a5305.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0465/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 465, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "24", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "move the green object from the right side of the burner into the silver pot and then remove it", "subtask": "move the green object from the right side of the burner into the silver pot and then remove it", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5d10f6284b38b7238966ff9.png"}, {"type": "image", "uri": "media/bdv2fail/b7af6187125aa804960e2271.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0466/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 466, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "put the blue fork on the upper right corner of the burner", "subtask": "put the blue fork on the upper right corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/858d719c2d354107c5d56c7a.png"}, {"type": "image", "uri": "media/bdv2fail/0a44be1c6ddf79debf9727a2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0467/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 467, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "take banana out of colander", "subtask": "take banana out of colander", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/12efcba41e6bda4f422e27b8.png"}, {"type": "image", "uri": "media/bdv2fail/e5414c9fb0fa5c87e4212793.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0468/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 468, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "green cloth had been moved to the top right side", "subtask": "green cloth had been moved to the top right side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/93a080f9c2761a5fc61ba8ba.png"}, {"type": "image", "uri": "media/bdv2fail/e44eaa2d560096fe3a65fe21.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0469/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 469, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the grey toy spatula to the middle of the stove on top of the dial.", "subtask": "move the grey toy spatula to the middle of the stove on top of the dial.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a31fb74daa73a2fb8e174500.png"}, {"type": "image", "uri": "media/bdv2fail/87cba07974b47a294654043f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0470/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 470, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "picking up the pot and putting it on the yellow towel.", "subtask": "picking up the pot and putting it on the yellow towel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eefd2a61d33f23500a58a79e.png"}, {"type": "image", "uri": "media/bdv2fail/d16b960af03243034ee49a8f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0471/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_99_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 471, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_99_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth to the top of the table", "subtask": "move the cloth to the top of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3cd12630ab5252b9769291f4.png"}, {"type": "image", "uri": "media/bdv2fail/218df4d9b5ed37da1cfbe488.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0472/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_01_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 472, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_01_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove the strawberry from the left burner", "subtask": "remove the strawberry from the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d30105b4d8a7e1683b41dda8.png"}, {"type": "image", "uri": "media/bdv2fail/4dc59863319ca455b3023b83.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0473/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_72_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 473, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_72_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "take the lion toy and place it below the drawer", "subtask": "take the lion toy and place it below the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/acd0111bd5e8882b17aef9c0.png"}, {"type": "image", "uri": "media/bdv2fail/df99b5cd92ae3cabefe1b28a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0474/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/41", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 474, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "41", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the cloth to right from left", "subtask": "move the cloth to right from left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}, {"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0475/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 475, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "folding the cloth from bottom right to the top left", "subtask": "folding the cloth from bottom right to the top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5b0bf7ffe77d8120a625944a.png"}, {"type": "image", "uri": "media/bdv2fail/d26e8a21d9436dd454a083b2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0477/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 477, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the red object on the left burner", "subtask": "put the red object on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9475a0acc7afe4c6013d89ef.png"}, {"type": "image", "uri": "media/bdv2fail/666e9dd5223127cbadac7991.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0478/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_04_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 478, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_04_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the red bell pepper to the yellow knife.", "subtask": "move the red bell pepper to the yellow knife.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/412a2e87e5db6f7db95f7a13.png"}, {"type": "image", "uri": "media/bdv2fail/debb5383eabe0f80cd09fb46.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0479/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 479, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_26_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the red pepper to the front left corner of the stove.", "subtask": "move the red pepper to the front left corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7e39879c4ac02d36882eeb3d.png"}, {"type": "image", "uri": "media/bdv2fail/8b4d481b8551e15402d0fa7e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0480/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_02_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 480, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_02_train", "episode_id": "0", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the eggplant to the pot", "subtask": "put the eggplant to the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e71541678506534af88b5816.png"}, {"type": "image", "uri": "media/bdv2fail/26d4a79f77d1865e2e5cf3e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0481/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_42_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 481, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_42_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the blue fork on top of the green cloth", "subtask": "put the blue fork on top of the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52500fffe2f46c52edcf994a.png"}, {"type": "image", "uri": "media/bdv2fail/d820b83f741ed95c6532c751.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0482/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 482, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the red spoon on the stove.", "subtask": "put the red spoon on the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9b90dfd3c0b69edbcb498f2.png"}, {"type": "image", "uri": "media/bdv2fail/d62a13dca5f3b8c3811ba8d8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0483/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 483, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "open the wooden drawer", "subtask": "open the wooden drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a53dba92f381f7006445f5bb.png"}, {"type": "image", "uri": "media/bdv2fail/57389ef7e821235fed6b1656.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0484/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 484, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put the stainless pot in the pot on the left burner", "subtask": "put the stainless pot in the pot on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2782fa0fe37141510589a3d9.png"}, {"type": "image", "uri": "media/bdv2fail/7b221eba5208bb04a79c1ab3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0485/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 485, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove spatula from plate sink", "subtask": "remove spatula from plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/204d280495a826912b8bed64.png"}, {"type": "image", "uri": "media/bdv2fail/b9f3ffe0b1b248aeee46759a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0486/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 486, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the upper right of the burner.", "subtask": "remove the silver pot from the upper right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fd1dbf5eb2689da071e02103.png"}, {"type": "image", "uri": "media/bdv2fail/eac5862f417c6f70b855240b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0487/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_44_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 487, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_44_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the blue figure on the top edge of the cloth", "subtask": "put the blue figure on the top edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2b8e2330d38c1563c9ac38e8.png"}, {"type": "image", "uri": "media/bdv2fail/dc4228d0728db133a5314b33.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0488/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 488, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "31", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove the chicken leg from the silver pot", "subtask": "remove the chicken leg from the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/605c2698db40b23da2e36801.png"}, {"type": "image", "uri": "media/bdv2fail/2f1425179ff6ab0dce75cc29.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0489/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 489, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "take the gray rabbit toy and put it on the left burner.", "subtask": "take the gray rabbit toy and put it on the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a9d7aa88a571d3824cc66924.png"}, {"type": "image", "uri": "media/bdv2fail/e7bcb9dd50b777b9601f76b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0490/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 490, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the yellow dish drying rack to the sink from the plastic pink cup", "subtask": "put the yellow dish drying rack to the sink from the plastic pink cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c583f095c9071d4e3bca7aa.png"}, {"type": "image", "uri": "media/bdv2fail/ed8be19109dc9a2b044da7aa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0491/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 491, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "put the silver pot to the lower right of the burner", "subtask": "put the silver pot to the lower right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f724029b5c75f7c9c4c51fd.png"}, {"type": "image", "uri": "media/bdv2fail/21a729f25895416530af561d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0492/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 492, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the orange towel to the left of the stove.", "subtask": "move the orange towel to the left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/436dd5a3bb58b32a994ec871.png"}, {"type": "image", "uri": "media/bdv2fail/4f9f0a8faf67213d382c824e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0493/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_03_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 493, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_03_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the blue cloth to the top of the stove", "subtask": "move the blue cloth to the top of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/817729855a101b7d59c24536.png"}, {"type": "image", "uri": "media/bdv2fail/e0b556772580d5bc06d71731.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0494/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 494, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the half orange in the silver", "subtask": "put the half orange in the silver", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/113282c5a87d7796e1031a58.png"}, {"type": "image", "uri": "media/bdv2fail/e6c67c205c9f478ef5da1c0c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0495/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 495, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "place the chicken on the orange cloth at the right upper side of the burner", "subtask": "place the chicken on the orange cloth at the right upper side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/07ad25d22f52e46bc5434c62.png"}, {"type": "image", "uri": "media/bdv2fail/2c37cdba113dd1d5fc4749f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0496/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_11_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 496, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_11_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "take the orange towel to the yellow plastic toy corn.", "subtask": "take the orange towel to the yellow plastic toy corn.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6740ab92c2bbc55198313fcd.png"}, {"type": "image", "uri": "media/bdv2fail/ab335c7f24fe59428a940c30.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0497/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 497, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move silver pot onto the right burner", "subtask": "move silver pot onto the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2fc973adb787cd5ebf1b48dd.png"}, {"type": "image", "uri": "media/bdv2fail/14370880da5faf7909bfac78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0498/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 498, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the cloth to the right of the pot.", "subtask": "move the cloth to the right of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e518077a57f79b81c7097538.png"}, {"type": "image", "uri": "media/bdv2fail/dc0da7265154d808b6b8fd75.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0499/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_06_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 499, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_06_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove the stainless pot in the pot from the left burner", "subtask": "remove the stainless pot in the pot from the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b221eba5208bb04a79c1ab3.png"}, {"type": "image", "uri": "media/bdv2fail/2782fa0fe37141510589a3d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0500/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 500, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow object to bottom right of the burner", "subtask": "move the yellow object to bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/af4a537373c5ed39f0623d5f.png"}, {"type": "image", "uri": "media/bdv2fail/fc35296f74fa3cbeb2cfbe71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0501/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 501, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put eggplant on plate", "subtask": "put eggplant on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/521abf6e09490588cc42d896.png"}, {"type": "image", "uri": "media/bdv2fail/df54d632021295075ba06308.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0502/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 502, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the brush to the upper middle of the table", "subtask": "move the brush to the upper middle of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7bc7090549ec933dfdb204cc.png"}, {"type": "image", "uri": "media/bdv2fail/7e4daddc2420837251ec2e47.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0503/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_15_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 503, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_15_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the napkin on the bottom right of the stove.", "subtask": "put the napkin on the bottom right of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d426044ecd9e4dfa01e60abf.png"}, {"type": "image", "uri": "media/bdv2fail/3e4f07afcf7b26a3ead73cdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0504/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 504, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth to the top left of the table from the center of the table", "subtask": "move the cloth to the top left of the table from the center of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/31408340bcbd3396c7ed46c8.png"}, {"type": "image", "uri": "media/bdv2fail/f264c92cc0b632c9951c7769.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0505/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_02_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 505, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_02_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move the pot to the yellow spatula", "subtask": "move the pot to the yellow spatula", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}, {"type": "image", "uri": "media/bdv2fail/1cb02cfb13d4475d5012d2f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0506/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 506, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the cloth from left to right.", "subtask": "move the cloth from left to right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/97de6df2eb6720e14986adb1.png"}, {"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0507/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 507, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the can at the right side of the burner", "subtask": "move the can at the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc01f813c13beb6cbcd794e6.png"}, {"type": "image", "uri": "media/bdv2fail/43a134743df93288cc3e3c78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0508/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 508, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9fbbca24a8a9fbb667d1f19c.png"}, {"type": "image", "uri": "media/bdv2fail/5a3f2d55fd39673d4e40df96.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0509/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 509, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "put the rice ball to the upper right side of the stove", "subtask": "put the rice ball to the upper right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f7e28b5ff3bdd992da8300dd.png"}, {"type": "image", "uri": "media/bdv2fail/a507cbaa3568c405e61f4525.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0510/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_13_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 510, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_13_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move silver pot to back right corner of stove top", "subtask": "move silver pot to back right corner of stove top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc5c86687fd9524e7c522f98.png"}, {"type": "image", "uri": "media/bdv2fail/e68ad6b23e94ea0074e1d640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0511/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_103_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 511, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_103_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot to the right burner", "subtask": "move the silver pot to the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/19c544696cb9a576ae92a44f.png"}, {"type": "image", "uri": "media/bdv2fail/d09a1e23e7d7f9f475be7c34.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0512/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_19_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 512, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_19_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the cloth to the top left", "subtask": "move the cloth to the top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b3cd04596cbe0743caf5261.png"}, {"type": "image", "uri": "media/bdv2fail/494f09dbce0909328827e2b0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0513/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 513, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the blue fork in the bottom right of the burner", "subtask": "put the blue fork in the bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6e02e143d34bae3641ab9abe.png"}, {"type": "image", "uri": "media/bdv2fail/45bfa177faba36b7c7453f0e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0514/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_19_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 514, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_19_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue brush to the upper side of the toy sink next to the yellow cloth.", "subtask": "move the blue brush to the upper side of the toy sink next to the yellow cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0fbf5aaeb262929ab97488be.png"}, {"type": "image", "uri": "media/bdv2fail/4ee3570fad61e966def81307.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0515/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 515, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "fold the cloth from left to right", "subtask": "fold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d8b2e12e74e93a6cd74ca867.png"}, {"type": "image", "uri": "media/bdv2fail/d8b2e12e74e93a6cd74ca867.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0516/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 516, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train", "episode_id": "4", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom left to top right", "subtask": "fold the cloth from bottom left to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3586851ccd5487351efa0461.png"}, {"type": "image", "uri": "media/bdv2fail/f12f21989ec916701bd4019d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0517/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_16_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 517, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_16_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the silver lid on the silver pot", "subtask": "put the silver lid on the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4aacf097b9ddc24a73c2efa6.png"}, {"type": "image", "uri": "media/bdv2fail/b294496cb738b6b7e33678d5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0518/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_22_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 518, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_22_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the orange cloth on the left stove", "subtask": "put the orange cloth on the left stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1952198e9c8e29d312d672f2.png"}, {"type": "image", "uri": "media/bdv2fail/6eba7c890cadea21880a9e77.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0519/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 519, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the silver pan on the upper left side of stove", "subtask": "put the silver pan on the upper left side of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/112623062501ef837e66e25a.png"}, {"type": "image", "uri": "media/bdv2fail/2f947d4d4c049667755a07e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0521/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 521, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove the corn from the upper right side of the burner", "subtask": "remove the corn from the upper right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e6fd821f7170f246601b6de3.png"}, {"type": "image", "uri": "media/bdv2fail/05f00798810589c29fe22dd1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0522/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_79_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 522, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_79_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth from the top left of the table to the bottom left", "subtask": "move the cloth from the top left of the table to the bottom left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3c00ab61e417dfcc122bd2fc.png"}, {"type": "image", "uri": "media/bdv2fail/f92c2170a61483ff8b412b20.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0523/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 523, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the little box to the top left side of the stove", "subtask": "move the little box to the top left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1af61238df201834801aaf5f.png"}, {"type": "image", "uri": "media/bdv2fail/b7736924a83e75ea82de41f8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0524/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 524, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train", "episode_id": "19", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the lid on the brown pot", "subtask": "put the lid on the brown pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/34d0128fad3c66f26c4e8fe9.png"}, {"type": "image", "uri": "media/bdv2fail/d39bcfec2d4b3e701c515546.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0525/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 525, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue cloth to the upper part of the stove", "subtask": "move the blue cloth to the upper part of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f25774d0c4ca9ff203f4981.png"}, {"type": "image", "uri": "media/bdv2fail/9f25774d0c4ca9ff203f4981.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0526/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 526, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "9", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the green object into the silver pot", "subtask": "move the green object into the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}, {"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0527/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 527, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "take the green sponge and move it on the upper left side of the stove.", "subtask": "take the green sponge and move it on the upper left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/67b970b3f09bc7683716b71b.png"}, {"type": "image", "uri": "media/bdv2fail/de88608fdd196c1e1e023139.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0529/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_75_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 529, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_75_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the blue towel to the upper right corner of the stove.", "subtask": "move the blue towel to the upper right corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b01bba73709b7ae77c93a420.png"}, {"type": "image", "uri": "media/bdv2fail/75a7d6d204701ba42c831d11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0530/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 530, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the pot to the upper right side of the left burner", "subtask": "move the pot to the upper right side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e3a28da7f36b35be4bd1250c.png"}, {"type": "image", "uri": "media/bdv2fail/4ffc97787b0abff950628d15.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0531/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 531, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "close tthe drawer", "subtask": "close tthe drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc32637edbb787fb9de025ed.png"}, {"type": "image", "uri": "media/bdv2fail/bc32637edbb787fb9de025ed.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0532/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 532, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_09_train", "episode_id": "11", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the red tomato to the back right corner of stove above cloth.", "subtask": "move the red tomato to the back right corner of stove above cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec08ffda626cede8d3d31f6e.png"}, {"type": "image", "uri": "media/bdv2fail/6927e1e015956cde7b4f8004.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0533/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_09_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 533, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_09_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the blue fork to the bottom right side of the table", "subtask": "move the blue fork to the bottom right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5bea955428c942b4b93d0289.png"}, {"type": "image", "uri": "media/bdv2fail/128415e9d099193d3276e4e0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0534/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 534, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the towel to the top side of the left burner.", "subtask": "move the towel to the top side of the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4e564177ef88ca6e8a84884.png"}, {"type": "image", "uri": "media/bdv2fail/e15609f091788e8448e94b2d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0535/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_02_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 535, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_02_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put spatula next to but not on plate sink", "subtask": "put spatula next to but not on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9f3ffe0b1b248aeee46759a.png"}, {"type": "image", "uri": "media/bdv2fail/204d280495a826912b8bed64.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0536/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 536, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "open the wooden box", "subtask": "open the wooden box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e7818742f29b833bbe529b26.png"}, {"type": "image", "uri": "media/bdv2fail/682fa35cf314c6a8ad3261b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0537/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_14_train/39", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 537, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_14_train", "episode_id": "39", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "remove the pot from the blue cloth.", "subtask": "remove the pot from the blue cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac72daba427d0cc5f6b9cdef.png"}, {"type": "image", "uri": "media/bdv2fail/ac72daba427d0cc5f6b9cdef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0538/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_68_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 538, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_68_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/71afa049da34579b6f777c6d.png"}, {"type": "image", "uri": "media/bdv2fail/a2243103f0ae3f591507eb71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0539/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_15_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 539, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_15_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the toy microwave", "subtask": "move the toy microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/111fc4b791dad864853861ed.png"}, {"type": "image", "uri": "media/bdv2fail/90ddf693fad1497e361eb499.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0540/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 540, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the green cloth from left to right and place it on the black faucet", "subtask": "move the green cloth from left to right and place it on the black faucet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d5da668f57dbae50dbca495.png"}, {"type": "image", "uri": "media/bdv2fail/9fd3b47bcf9d897c052e765e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0541/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 541, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the red rectangular block below the ice cream toy", "subtask": "move the red rectangular block below the ice cream toy", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/38407db3eed87731a5ea2399.png"}, {"type": "image", "uri": "media/bdv2fail/b1a316f25cf30f886e3b4547.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0542/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/29", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 542, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "29", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the orange towel close by but not on the lower part of stove", "subtask": "move the orange towel close by but not on the lower part of stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6f1bdfa08248426409a98767.png"}, {"type": "image", "uri": "media/bdv2fail/35511d4afd2354679282cf24.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0543/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 543, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "put the blue toy fork to the upper right burner from the upper left burner.", "subtask": "put the blue toy fork to the upper right burner from the upper left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e82556077fc148d1ff24724f.png"}, {"type": "image", "uri": "media/bdv2fail/57ed1961ac71af6086f633bd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0544/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 544, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "unfold the cloth from right to left", "subtask": "unfold the cloth from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e6128aff8323ad88f97404a.png"}, {"type": "image", "uri": "media/bdv2fail/3e6128aff8323ad88f97404a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0545/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 545, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "11", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the red tomatoes basket from counter to sink", "subtask": "put the red tomatoes basket from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/57578533979e054da972be29.png"}, {"type": "image", "uri": "media/bdv2fail/ed028fcec376804e6ce64226.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0546/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 546, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "24", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the spoon from the left to the right and put it in top of the green cloth.", "subtask": "move the spoon from the left to the right and put it in top of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/433a332fedef5d025d4d3bce.png"}, {"type": "image", "uri": "media/bdv2fail/bf69525ae0fc9e844c647e73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0547/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_28_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 547, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_28_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 249, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/36c7514b37d61f532f7eecaa.png"}, {"type": "image", "uri": "media/bdv2fail/c80e6d187249b89b9aaef38b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0548/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_14_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 548, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_14_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "pick up the red fork and place it on the purple napkin", "subtask": "pick up the red fork and place it on the purple napkin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9e5f3b8ac2b7b3daa965e7d2.png"}, {"type": "image", "uri": "media/bdv2fail/8de1c7a1c284a9ce76507edf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0549/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 549, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the green spatula to the right bottom side of the table.", "subtask": "move the green spatula to the right bottom side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a1f5ee1edc90360a984df94d.png"}, {"type": "image", "uri": "media/bdv2fail/1bfbfd8368f54b601bf91d36.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0550/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 550, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "put the knife on top of the yellow cloth", "subtask": "put the knife on top of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/09bee9da77127a3cee8970b6.png"}, {"type": "image", "uri": "media/bdv2fail/9ae0bfbde3bfbca5c48f4f61.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0551/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 551, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the can to the bottom left of the table", "subtask": "move the can to the bottom left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/360a78d99eafb054388efd9c.png"}, {"type": "image", "uri": "media/bdv2fail/88accd289dce509b5157514a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0552/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 552, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot to the left upper corner of the stove", "subtask": "move the silver pot to the left upper corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/802c93c13d43332030ff3374.png"}, {"type": "image", "uri": "media/bdv2fail/762b08c0b02ccb9dafe75dbb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0553/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_64_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 553, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_64_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the yellow cloth upward the stove", "subtask": "move the yellow cloth upward the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52938137d441f4189764d1d7.png"}, {"type": "image", "uri": "media/bdv2fail/52938137d441f4189764d1d7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0554/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 554, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_07_train", "episode_id": "20", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the silver to the right of burner", "subtask": "move the silver to the right of burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e2247496372030225d89f7d5.png"}, {"type": "image", "uri": "media/bdv2fail/2f5f6b1c11dc209d11d071ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0555/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_39_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 555, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_39_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the pot on the bottom right of the stove", "subtask": "put the pot on the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d426044ecd9e4dfa01e60abf.png"}, {"type": "image", "uri": "media/bdv2fail/3e4f07afcf7b26a3ead73cdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0556/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 556, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upper left side of the table", "subtask": "move the silver pot to the upper left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/99672f5be4f08bd9cc9845ed.png"}, {"type": "image", "uri": "media/bdv2fail/99672f5be4f08bd9cc9845ed.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0557/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 557, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the pot from the right upper part of the stove.", "subtask": "remove the pot from the right upper part of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a99bbadd50f5a1b1f94b3879.png"}, {"type": "image", "uri": "media/bdv2fail/10632eb0957c40efef4f0a39.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0558/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_02_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 558, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_02_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot from the burner to the top of a blue cloth", "subtask": "move the silver pot from the burner to the top of a blue cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/378b665f8484e78ed9d2fa29.png"}, {"type": "image", "uri": "media/bdv2fail/bffebe9b7211536c2dc094b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0559/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 559, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "remove the lid from the silver bowl", "subtask": "remove the lid from the silver bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ee7766f25a37ddd4e16518e.png"}, {"type": "image", "uri": "media/bdv2fail/9ee7766f25a37ddd4e16518e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0560/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 560, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the green cloth to the bowl", "subtask": "move the green cloth to the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a375d4028dcc33f4bb9e969c.png"}, {"type": "image", "uri": "media/bdv2fail/bbadef55695967a32a6881f8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0561/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 561, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "grab the yellow toy and move it to the top of the left burner", "subtask": "grab the yellow toy and move it to the top of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2499442e374b7fef3489b28c.png"}, {"type": "image", "uri": "media/bdv2fail/a88b48fb5b46b7cd1d78475a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0562/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_64_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 562, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_64_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the cloth to right from left", "subtask": "move the cloth to right from left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}, {"type": "image", "uri": "media/bdv2fail/97de6df2eb6720e14986adb1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0563/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 563, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue fork to the green towel", "subtask": "move the blue fork to the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aa67636d2834cd9744c6e590.png"}, {"type": "image", "uri": "media/bdv2fail/a9da7a85afe8afa40602f3b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0564/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 564, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the blue towel to the top side of the faucet.", "subtask": "move the blue towel to the top side of the faucet.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9189bbaba8039f2d7f3760b9.png"}, {"type": "image", "uri": "media/bdv2fail/fa20c9e1cf504b31724106c2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0565/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 565, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put cup close to but not on plate", "subtask": "put cup close to but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d70567d9bf2c993a715fe709.png"}, {"type": "image", "uri": "media/bdv2fail/3677a790ecb1ad38d831c2e2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0566/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 566, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "24", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the silver pot on the yellow cloth", "subtask": "put the silver pot on the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/439f1fdf11263d35555de211.png"}, {"type": "image", "uri": "media/bdv2fail/0c7e445be1566630533750f6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0567/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 567, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the spatula and place it just behind the orange peel.", "subtask": "move the spatula and place it just behind the orange peel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79e26cf73d134fe9b9181273.png"}, {"type": "image", "uri": "media/bdv2fail/d7260464af59a8e5e633d2cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0568/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 568, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the metal can to the right of the stove", "subtask": "move the metal can to the right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0c202b8368aa2e97b707f9f2.png"}, {"type": "image", "uri": "media/bdv2fail/f45fcb652f738795c30a5f78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0569/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 569, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the cloth with a red circle from right to left", "subtask": "move the cloth with a red circle from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/924374499e8937a81bdd5a4a.png"}, {"type": "image", "uri": "media/bdv2fail/74681f140d9c5ea2361a859f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0571/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 571, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train", "episode_id": "19", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "fold the cloth from top left to bottom right", "subtask": "fold the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d6b3943d6e632f0785889abc.png"}, {"type": "image", "uri": "media/bdv2fail/5460ca7d4614f650b855bc47.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0572/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_19_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 572, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_19_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the silver pot on the stove aer", "subtask": "put the silver pot on the stove aer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3234185a0bfa97411435a2a5.png"}, {"type": "image", "uri": "media/bdv2fail/000348d821b8723baadab927.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0573/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 573, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the strawberry underneath the silver pan from the left stove burner.", "subtask": "remove the strawberry underneath the silver pan from the left stove burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/45e19384626ede0e6d448684.png"}, {"type": "image", "uri": "media/bdv2fail/99a0a67e874e94453c0d7f3a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0574/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_09_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 574, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_09_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold the cloth from left to right", "subtask": "fold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/71a424a78a275d8c6367c71c.png"}, {"type": "image", "uri": "media/bdv2fail/dcfff03c8ef9d67e0ef9ea70.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0575/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 575, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the green cloth to the bottom left corner of the stove", "subtask": "move the green cloth to the bottom left corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a38afdf6ee867f5d8a77f241.png"}, {"type": "image", "uri": "media/bdv2fail/dc3a4c42331e380a39ced400.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0576/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_110_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 576, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_110_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the yellow spatula to back right corner of stove top.", "subtask": "move the yellow spatula to back right corner of stove top.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/80ae2d3150b7a5723e0f8276.png"}, {"type": "image", "uri": "media/bdv2fail/522ed9bfbb3f41aaed804c19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0577/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_69_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 577, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_69_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put beet in pot sink", "subtask": "put beet in pot sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9a51ebbab1bc853d30bc1681.png"}, {"type": "image", "uri": "media/bdv2fail/252fddca8057992971cc65ac.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0578/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/29", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 578, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "29", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf99c287245707462d9c9a16.png"}, {"type": "image", "uri": "media/bdv2fail/c990e8c0e97d6dc50dfc9b5f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0579/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 579, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "take blueberries off plate sink", "subtask": "take blueberries off plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/550c7c3f0f63ad47d813d1df.png"}, {"type": "image", "uri": "media/bdv2fail/91d19caca17f4862a0025b59.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0580/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_blueberries_off_plate_sink_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 580, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_blueberries_off_plate_sink_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "take the blue cloth to the left side of the burner from the right.", "subtask": "take the blue cloth to the left side of the burner from the right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1220b0cd25acbee2da4bebc8.png"}, {"type": "image", "uri": "media/bdv2fail/50b596b2e642d75615e7fccd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0581/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 581, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "pick up the silver pot and put it at the opposite corner of the stove", "subtask": "pick up the silver pot and put it at the opposite corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6c04c9a4660d662a4d09965b.png"}, {"type": "image", "uri": "media/bdv2fail/0cfe07136d1f07f9bc3de241.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0582/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 582, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "3", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "put the silver pot lid on the top of the silver pot", "subtask": "put the silver pot lid on the top of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/58e7d8debd78055fa8271394.png"}, {"type": "image", "uri": "media/bdv2fail/e1362c5991ab6987d106102f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0583/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 583, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the yellow doll on the red tomato", "subtask": "put the yellow doll on the red tomato", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/223ec694abb51b4ba5e33564.png"}, {"type": "image", "uri": "media/bdv2fail/64bd04c75ab8efcf838951dc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0584/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 584, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train", "episode_id": "24", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the cloth to the left bottom side of the cooker from the top right side", "subtask": "move the cloth to the left bottom side of the cooker from the top right side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6fe0c0bb254d0d639d9420bd.png"}, {"type": "image", "uri": "media/bdv2fail/a440eaafa8ed44241d69ebf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0585/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 585, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upper left edge of the stove.", "subtask": "move the silver pot to the upper left edge of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/37fbe2530f19896e16872ecd.png"}, {"type": "image", "uri": "media/bdv2fail/8599a6a4a54f0368f2d3c6e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0586/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 586, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put cup to drying rack from sink", "subtask": "put cup to drying rack from sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f0350fc1da496a4bb9b0fc6d.png"}, {"type": "image", "uri": "media/bdv2fail/97a903e03676c4e3aea53ff2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0587/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 587, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the silver pot to the left top corner of the burner", "subtask": "move the silver pot to the left top corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3414ba57fb6a578a86e319af.png"}, {"type": "image", "uri": "media/bdv2fail/c64fd5a383884dc0495eda73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0589/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 589, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the white cloth to top edge of the burner from bottom edge of the burner", "subtask": "move the white cloth to top edge of the burner from bottom edge of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/11a24e351de4b94d56db2c44.png"}, {"type": "image", "uri": "media/bdv2fail/2702e87ff3dc7d7ea98d6801.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0590/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 590, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "remove banana from colander", "subtask": "remove banana from colander", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6edb1091c31cf813cc662ed3.png"}, {"type": "image", "uri": "media/bdv2fail/ac5998ec34d707221cff590a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0591/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train/32", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 591, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train", "episode_id": "32", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "unfold the cloth from left to right", "subtask": "unfold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/74c518c19dc064e5d76a3fe4.png"}, {"type": "image", "uri": "media/bdv2fail/74c518c19dc064e5d76a3fe4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0592/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_12_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 592, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_12_train", "episode_id": "4", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the green fork to right of the silver pot", "subtask": "move the green fork to right of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4444af3ac5316c8102f15350.png"}, {"type": "image", "uri": "media/bdv2fail/ddd74a1844da9611a7fbf8b5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0593/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 593, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the yellow cloth on the right side of the table.", "subtask": "put the yellow cloth on the right side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/93cc244b5dd9b6451709acd1.png"}, {"type": "image", "uri": "media/bdv2fail/73022d869b1beefe4cda3405.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0594/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_28_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 594, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_28_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove a blue stuffed animal from the drawer.", "subtask": "remove a blue stuffed animal from the drawer.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/820d263ae01a56b5face038e.png"}, {"type": "image", "uri": "media/bdv2fail/fd83b81123d86595943ccc8d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0595/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 595, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the broccoli to the middle of the table", "subtask": "move the broccoli to the middle of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e2e8bf8a73cac16611fff8af.png"}, {"type": "image", "uri": "media/bdv2fail/34a0ded74dbcbee0972c414d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0596/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 596, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the brown lid on the black pot", "subtask": "put the brown lid on the black pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e15ddef653eba328a364d00.png"}, {"type": "image", "uri": "media/bdv2fail/74934245d7b66974af3ff857.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0597/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 597, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the capsicum in the black pot", "subtask": "place the capsicum in the black pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/96837ab6c4383dc63e004864.png"}, {"type": "image", "uri": "media/bdv2fail/4c107d8927d96817d4e73b39.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0598/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_104_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 598, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_104_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the yellow duck from the lower left side of the table", "subtask": "remove the yellow duck from the lower left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0b21c28c9f72143dc91fc6c9.png"}, {"type": "image", "uri": "media/bdv2fail/4cf5bc8939d12ababaa5116f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0599/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train/35", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 599, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train", "episode_id": "35", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth to the right side of the stove", "subtask": "move the cloth to the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ce2ab4922adfc005e76b76a3.png"}, {"type": "image", "uri": "media/bdv2fail/89c7a1c81e063b4eef1f0774.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0600/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_03_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 600, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_03_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put cup to counter from sink", "subtask": "put cup to counter from sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e5db339e69a1888477da9d9f.png"}, {"type": "image", "uri": "media/bdv2fail/9ef2581e8a438faa509fdffc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0601/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 601, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "remove the napkin from the bottom right of the stove.", "subtask": "remove the napkin from the bottom right of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e4f07afcf7b26a3ead73cdc.png"}, {"type": "image", "uri": "media/bdv2fail/d426044ecd9e4dfa01e60abf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0602/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 602, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the chocolate bar to the bottom of the left burner.", "subtask": "move the chocolate bar to the bottom of the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/275543f16e4856019bd6e0b5.png"}, {"type": "image", "uri": "media/bdv2fail/cee345aef808e5192b82a4f4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0603/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 603, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "place the brush on the yellow cloth", "subtask": "place the brush on the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/37c4dca1fb8c1328673f0976.png"}, {"type": "image", "uri": "media/bdv2fail/81860e40dd2ca4ba1b54c7a1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0604/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_10_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 604, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_10_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the carrot to the top side of the table.", "subtask": "move the carrot to the top side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dc23529253d7e8c92a258452.png"}, {"type": "image", "uri": "media/bdv2fail/0102cb17f61e200d318d78cc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0605/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_18_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 605, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_18_train", "episode_id": "31", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the steel pot close by but not on the left burner", "subtask": "put the steel pot close by but not on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4748fc852e7a5980f612b5f.png"}, {"type": "image", "uri": "media/bdv2fail/415ab6ceaf266d9719e4107b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0606/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_25_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 606, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_25_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth to the center from top left of the white wall", "subtask": "move the cloth to the center from top left of the white wall", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c3b618e18d13e3431ee8bba4.png"}, {"type": "image", "uri": "media/bdv2fail/c880ccf918024d434d07cb4a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0607/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_02_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 607, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_02_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "remove the yellow knife from the purple cloth", "subtask": "remove the yellow knife from the purple cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a43a55dffd7a16bce3703112.png"}, {"type": "image", "uri": "media/bdv2fail/109387462a452ed28fa71e6b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0608/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_42_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 608, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_42_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the corn toy yellow toy on the orange sponge", "subtask": "put the corn toy yellow toy on the orange sponge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2582749ce00c830a17349954.png"}, {"type": "image", "uri": "media/bdv2fail/8acbf18fd640c27d7a64daa7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0609/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 609, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "remove blueberries from plate sink", "subtask": "remove blueberries from plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cb2ebef068c027d40732b148.png"}, {"type": "image", "uri": "media/bdv2fail/cb2ebef068c027d40732b148.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0610/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 610, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "12", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom rigth to up left", "subtask": "fold the cloth from bottom rigth to up left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a559187d192891f7cead6612.png"}, {"type": "image", "uri": "media/bdv2fail/88867b98ae1a149ea32994ca.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0611/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_19_train/29", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 611, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_19_train", "episode_id": "29", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the can at the left side of the burner", "subtask": "move the can at the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc01f813c13beb6cbcd794e6.png"}, {"type": "image", "uri": "media/bdv2fail/43a134743df93288cc3e3c78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0612/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 612, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "moving the spatula from the center bottom of two burner to center top", "subtask": "moving the spatula from the center bottom of two burner to center top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/78912662b490240b07844e31.png"}, {"type": "image", "uri": "media/bdv2fail/ce5918e237f8e51bae7fbb1f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0613/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 613, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the bowl from the back left to the front right corner.", "subtask": "move the bowl from the back left to the front right corner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bffc7ba1b7cc212d6ed0a846.png"}, {"type": "image", "uri": "media/bdv2fail/ec02bd65b6583fd4c5d4b94a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0614/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 614, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the yellow cloth", "subtask": "remove the silver pot from the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/00a7285242f1286e37a75331.png"}, {"type": "image", "uri": "media/bdv2fail/0d01670dcbeb903c6f15302d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0615/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 615, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the egg to the bottom right corner of the kitchen", "subtask": "move the egg to the bottom right corner of the kitchen", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/67acc55971616a0aef8883e4.png"}, {"type": "image", "uri": "media/bdv2fail/54171914038640c7a2485042.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0616/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_05_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 616, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_05_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the red fork to the top right of the stove", "subtask": "move the red fork to the top right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/218a2553401dde9b58b4767c.png"}, {"type": "image", "uri": "media/bdv2fail/b5be400795390358a2b8d06a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0617/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 617, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "pick up the red spatula and place it above the right burner.", "subtask": "pick up the red spatula and place it above the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/16d516b28fb0006d8c2e519c.png"}, {"type": "image", "uri": "media/bdv2fail/f48a4c7c67434dc82eace01a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0618/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 618, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upper side of the blue cloth", "subtask": "move the silver pot to the upper side of the blue cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9ce29d894602fc9b44bea0e.png"}, {"type": "image", "uri": "media/bdv2fail/714d552e8f8084419258beb6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0619/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_108_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 619, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_108_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the knife from top of the yellow cloth", "subtask": "remove the knife from top of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ae0bfbde3bfbca5c48f4f61.png"}, {"type": "image", "uri": "media/bdv2fail/09bee9da77127a3cee8970b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0620/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 620, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the yellow cloth from the middle front of the burners to behind the left side of the burners.", "subtask": "move the yellow cloth from the middle front of the burners to behind the left side of the burners.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b13a7904f6d38c43dac100f.png"}, {"type": "image", "uri": "media/bdv2fail/120004508eca18c9559e4284.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0621/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 621, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 255, "role": "outcome"}}
{"state": {"task": "move the silver bowl to the lower side of the table.", "subtask": "move the silver bowl to the lower side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fc170c5286f9afa88eed2630.png"}, {"type": "image", "uri": "media/bdv2fail/74fd4002e211caa8d7b6ab98.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0622/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 622, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "unfold the cloth from bottom to top", "subtask": "unfold the cloth from bottom to top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/edb6e229a84aa3829ad0e58c.png"}, {"type": "image", "uri": "media/bdv2fail/5ff78d5e703fae8449a20ad3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0623/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 623, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "cover the metal clamp with the pot.", "subtask": "cover the metal clamp with the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4fba0a9455049f1be4923ad8.png"}, {"type": "image", "uri": "media/bdv2fail/f2f20c1afa362c069a4dafdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0624/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 624, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put silver pot at the bottom right side of the burner near the left switch", "subtask": "put silver pot at the bottom right side of the burner near the left switch", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eb090f9d4f08ae111b1598d7.png"}, {"type": "image", "uri": "media/bdv2fail/266a0024b2e61703d1eafbf9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0625/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 625, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the cloth from left to right", "subtask": "move the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/db5d032dab6631088ee96ee0.png"}, {"type": "image", "uri": "media/bdv2fail/94b20ac30d3e2f4726752c5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0626/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 626, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the pink bunny to the left side of the table", "subtask": "move the pink bunny to the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/de82e16ecea5dead8155b6db.png"}, {"type": "image", "uri": "media/bdv2fail/84726fede6eb3074fa0263a0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0627/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_14_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 627, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_14_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1298afc119b4d6883b0d5e00.png"}, {"type": "image", "uri": "media/bdv2fail/1298afc119b4d6883b0d5e00.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0628/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 628, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train", "episode_id": "0", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "take the pineapple and put it below the right side burner", "subtask": "take the pineapple and put it below the right side burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/147dcb889901237225a374a9.png"}, {"type": "image", "uri": "media/bdv2fail/f54795d84aa70f82995f6419.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0630/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_21_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 630, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_21_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow knife between the burner and the sink", "subtask": "move the yellow knife between the burner and the sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b08094972edd79ade776c1e8.png"}, {"type": "image", "uri": "media/bdv2fail/18d057d5ab58fd5e66790e2e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0631/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_59_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 631, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_59_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot on the green cloth .", "subtask": "move the silver pot on the green cloth .", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/58d9e2d2462305fa33de9dc3.png"}, {"type": "image", "uri": "media/bdv2fail/a907e627968be5be43dc19bf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0632/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_40_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 632, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_40_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the bottom right of the stove", "subtask": "move the blue spoon to the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/355752f3145a544ae2844058.png"}, {"type": "image", "uri": "media/bdv2fail/355752f3145a544ae2844058.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0633/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 633, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "1", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "push the blue cloth to the upper left of the burner", "subtask": "push the blue cloth to the upper left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3977181724dce9d999d17843.png"}, {"type": "image", "uri": "media/bdv2fail/5e63ddc3c4c04ec528c27256.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0634/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 634, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "pick up the toolbox.", "subtask": "pick up the toolbox.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/24557c4a225f8c8920d0ff1a.png"}, {"type": "image", "uri": "media/bdv2fail/df7c432e5f62f58e82974b38.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0635/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 635, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_red_srewdriver_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put the orange cloth on the right stove", "subtask": "put the orange cloth on the right stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1952198e9c8e29d312d672f2.png"}, {"type": "image", "uri": "media/bdv2fail/6eba7c890cadea21880a9e77.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0636/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 636, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the red object from the top of the cloth to the table", "subtask": "move the red object from the top of the cloth to the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4099d5b9200d553d4785f374.png"}, {"type": "image", "uri": "media/bdv2fail/4099d5b9200d553d4785f374.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0637/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_07_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 637, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_07_train", "episode_id": "17", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the spatula to the middle of two red burners", "subtask": "move the spatula to the middle of two red burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d0619e759b4d75485b054e3c.png"}, {"type": "image", "uri": "media/bdv2fail/4ad0d389ab5a3a4636a08471.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0638/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 638, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the silver pot to the left side of the left burner", "subtask": "move the silver pot to the left side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cefff62a8d323345966d9699.png"}, {"type": "image", "uri": "media/bdv2fail/8e05d48f1dd34017ecbae070.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0639/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 639, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_05_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the yellow object on the center of orange cloth", "subtask": "put the yellow object on the center of orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e944b01f84d08ffd14e4bd0f.png"}, {"type": "image", "uri": "media/bdv2fail/d6f8e9ecd60e145188e72ff6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0640/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_51_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 640, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_51_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a46a657d90cfb88bc5dbc5ae.png"}, {"type": "image", "uri": "media/bdv2fail/9d4f3f805e9e2e9d6e60adcc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0642/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 642, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "move the green spatula to the right side of the left burner", "subtask": "move the green spatula to the right side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d68a7d8391478253271339d3.png"}, {"type": "image", "uri": "media/bdv2fail/6300092461b3fe105761e809.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0643/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 643, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_24_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the yellow banana to the top right of the stove", "subtask": "move the yellow banana to the top right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d0a0d4915cc35d19b06387c5.png"}, {"type": "image", "uri": "media/bdv2fail/21d559a5f7975187f6fbe080.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0644/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 644, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver lid of the pot to the left side of the table", "subtask": "move the silver lid of the pot to the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9e043498861d7cc6d34ffe61.png"}, {"type": "image", "uri": "media/bdv2fail/51007b3ff07b82bad8581a51.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0645/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 645, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "put cucumber in cup", "subtask": "put cucumber in cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5fbbaf7e9267c641569720e1.png"}, {"type": "image", "uri": "media/bdv2fail/dee962368aadabff544d193d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0646/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 646, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom left side of the burner", "subtask": "move the silver pot to the bottom left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5f07682fe159092e4303b92e.png"}, {"type": "image", "uri": "media/bdv2fail/805a964c8b3743fff89e5eee.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0647/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_83_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 647, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_83_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red bell pepper into bottom left of the burner", "subtask": "move the red bell pepper into bottom left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c829780dbb0b036c7ea0e29.png"}, {"type": "image", "uri": "media/bdv2fail/4270220db5bbf7711277da5e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0648/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_78_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 648, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_78_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "take cup off plate", "subtask": "take cup off plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4ddb2de4ef12775a25cb16c1.png"}, {"type": "image", "uri": "media/bdv2fail/4ac2dec75a353444bd959fd7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0649/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_cup_off_plate_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 649, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_cup_off_plate_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the purple towel to the lower right side of the burner.", "subtask": "move the purple towel to the lower right side of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6f09b245e433adf0e664da97.png"}, {"type": "image", "uri": "media/bdv2fail/1e63651c2af4ee80b6b74a25.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0650/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 650, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the pot to the top left of the stove.", "subtask": "move the pot to the top left of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec63d22963aa70f9b0318337.png"}, {"type": "image", "uri": "media/bdv2fail/2bb0eeccacc3bab266cab577.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0651/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 651, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the pot on the purple cloth.", "subtask": "move the pot on the purple cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0d16c26c272254622aad4c97.png"}, {"type": "image", "uri": "media/bdv2fail/cfd08aaae2c948c020367ed3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0652/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 652, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the yellow knife to inbetween the left burner and the can and put it from bottom of left burner", "subtask": "move the yellow knife to inbetween the left burner and the can and put it from bottom of left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2a8be9055f21608e88c47aff.png"}, {"type": "image", "uri": "media/bdv2fail/b7f4b69e7d09ab12b82766d8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0653/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 653, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 255, "role": "outcome"}}
{"state": {"task": "put the blue cup into the basket", "subtask": "put the blue cup into the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/618f735c173259d70592d0ef.png"}, {"type": "image", "uri": "media/bdv2fail/6863c54f5074103a8182f723.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0654/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/43", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 654, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "43", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the silver bowl on top of the orange cloth", "subtask": "move the silver bowl on top of the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5add9c5994df28b75f9d8c12.png"}, {"type": "image", "uri": "media/bdv2fail/ded2e2e1c49ef673d35b86da.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0655/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 655, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a46a657d90cfb88bc5dbc5ae.png"}, {"type": "image", "uri": "media/bdv2fail/9d4f3f805e9e2e9d6e60adcc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0657/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 657, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "move the metal bowl to the middle upper part of the table.", "subtask": "move the metal bowl to the middle upper part of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/29ffa4ac92ed3cee28963bcd.png"}, {"type": "image", "uri": "media/bdv2fail/59ccd025920406eddadf2f5b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0658/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 658, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the green towel from the bottom left to the top right of the burner.", "subtask": "move the green towel from the bottom left to the top right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6f5466de30f17f1ff35785f8.png"}, {"type": "image", "uri": "media/bdv2fail/7b36c36489b429f93624d2a6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0659/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_00_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 659, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_00_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "put the silver lid on the silver pot", "subtask": "put the silver lid on the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/12a3982a84be10dcc2289c44.png"}, {"type": "image", "uri": "media/bdv2fail/be32103799c792ede7b439c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0660/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 660, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the cheese to the right of the blue brush.", "subtask": "move the cheese to the right of the blue brush.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8ecd81906d7736b2c835391.png"}, {"type": "image", "uri": "media/bdv2fail/9f5941add05dd616c3c28968.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0661/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_91_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 661, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_91_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "place the silver bowl on the stove", "subtask": "place the silver bowl on the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e48fc69974c490a71851ca31.png"}, {"type": "image", "uri": "media/bdv2fail/54646820b26a879c52cb0706.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0662/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_37_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 662, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_37_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the towel to right side of the burner", "subtask": "move the towel to right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/acd9c3c471778f3cfb87bd42.png"}, {"type": "image", "uri": "media/bdv2fail/331eec395a9cedd102c4e5d4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0663/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_15_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 663, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_15_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put spoon in bowl sink", "subtask": "put spoon in bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc645d1faf0738f842573afc.png"}, {"type": "image", "uri": "media/bdv2fail/bc645d1faf0738f842573afc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0664/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 664, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "24", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "take the half cut tomato and put it inside the silver pot", "subtask": "take the half cut tomato and put it inside the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f23f867207d75062a2e456ba.png"}, {"type": "image", "uri": "media/bdv2fail/7e70bba9c906effb0f11d8cb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0665/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_97_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 665, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_97_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "take the silver bowl and put the red object (tomato) on top of it.", "subtask": "take the silver bowl and put the red object (tomato) on top of it.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/187ca733a71978cd371599da.png"}, {"type": "image", "uri": "media/bdv2fail/51cb73c834f95be41d557a8c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0666/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_39_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 666, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_39_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "put the blue fork in the bottom left of the burner", "subtask": "put the blue fork in the bottom left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6e02e143d34bae3641ab9abe.png"}, {"type": "image", "uri": "media/bdv2fail/45bfa177faba36b7c7453f0e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0667/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_19_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 667, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_19_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver with blue handle spoon to the left side of the purple towel", "subtask": "move the silver with blue handle spoon to the left side of the purple towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/05b563a522e43973b4af866f.png"}, {"type": "image", "uri": "media/bdv2fail/cf4da7c1d8f61b80a21df127.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0668/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 668, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the grey sink to the middle of the burner from the right burner.", "subtask": "move the grey sink to the middle of the burner from the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7557e94fe58cabe8eb5e0db5.png"}, {"type": "image", "uri": "media/bdv2fail/064933e5ed3615b4131fd62f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0669/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 669, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the silver screw from left corner of the table to the top of the table.", "subtask": "move the silver screw from left corner of the table to the top of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/080af74866c974dbf4e2604f.png"}, {"type": "image", "uri": "media/bdv2fail/2a0e138e25f7a006347ac757.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0670/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 670, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move the cloth from the left burner to the right burner", "subtask": "move the cloth from the left burner to the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0475b18118c518a6e66c64b7.png"}, {"type": "image", "uri": "media/bdv2fail/e9c678dba3c0355a6ece7fa3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0671/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 671, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move cloth from top to bottom", "subtask": "move cloth from top to bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/72d30aa3d34fb9a81eedb5e2.png"}, {"type": "image", "uri": "media/bdv2fail/fbd5256059f152bb7404ce4b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0672/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 672, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "fold the cloth from the bottom right to the left", "subtask": "fold the cloth from the bottom right to the left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0cdc2b9997b5be716c2de295.png"}, {"type": "image", "uri": "media/bdv2fail/1cfcb70b831c8c1b44e69b0d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0673/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_22_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 673, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_22_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6bb2fb3b3e435c8fe622c503.png"}, {"type": "image", "uri": "media/bdv2fail/aba7dfe4d7f485e002ab0ec2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0674/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 674, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "remove corn from bowl sink", "subtask": "remove corn from bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3124b5ddd6d73d5c63413db7.png"}, {"type": "image", "uri": "media/bdv2fail/a59d7176f4cfde0275bc3597.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0675/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 675, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put the blue spoon to the left of the hot dog", "subtask": "put the blue spoon to the left of the hot dog", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c41861b6f740f9b3288edc2b.png"}, {"type": "image", "uri": "media/bdv2fail/a6de1abe926b813e375e7204.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0676/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 676, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put cup in sink", "subtask": "put cup in sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/66610af9e667ac64d0de58e3.png"}, {"type": "image", "uri": "media/bdv2fail/33fe5830f6665ab65e9d49d6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0677/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/29", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 677, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "29", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the yellow spoon to the bottom left side of the stove.", "subtask": "move the yellow spoon to the bottom left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7c8840a5746827bb7986e84b.png"}, {"type": "image", "uri": "media/bdv2fail/838eff72ab2968abc857c4cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0678/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 678, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the blue spoon from the top of the green towel", "subtask": "remove the blue spoon from the top of the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5e256504f59abbfbc808e14.png"}, {"type": "image", "uri": "media/bdv2fail/c728ac183e0042e2196d76b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0679/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_31_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 679, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_31_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth to the right of table", "subtask": "move the cloth to the right of table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8d301e630c2556bf4fee6076.png"}, {"type": "image", "uri": "media/bdv2fail/966df0bd78e9781d454f5ac8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0680/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 680, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put cucumber in cup", "subtask": "put cucumber in cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dc5266cb77299e2ccd10ce9.png"}, {"type": "image", "uri": "media/bdv2fail/378ef75919fd6216f4a79b58.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0681/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 681, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put the green object towel in the silver pot", "subtask": "put the green object towel in the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a579474f3d1b86375cbf1b68.png"}, {"type": "image", "uri": "media/bdv2fail/f5009d123f10282ba5b8ac73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0682/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_74_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 682, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_74_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove the corn from the silver pot", "subtask": "remove the corn from the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8ea5130969d36de31f2e4cd.png"}, {"type": "image", "uri": "media/bdv2fail/f8ea5130969d36de31f2e4cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0683/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 683, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "9", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the yellow towel near but not on the left burner", "subtask": "put the yellow towel near but not on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70c5fe5722bd30d78ae38041.png"}, {"type": "image", "uri": "media/bdv2fail/bc459149bd15a4f79827d310.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0684/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 684, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the banana to the middle upper part of the table", "subtask": "move the banana to the middle upper part of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/29ffa4ac92ed3cee28963bcd.png"}, {"type": "image", "uri": "media/bdv2fail/59ccd025920406eddadf2f5b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0685/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 685, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put banana in colander", "subtask": "put banana in colander", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4cce6af132a4d9ef7946e6e3.png"}, {"type": "image", "uri": "media/bdv2fail/4cce6af132a4d9ef7946e6e3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0686/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 686, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_banana_in_colander_train", "episode_id": "4", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the cloth to the middle from the bottom edge of the table", "subtask": "move the cloth to the middle from the bottom edge of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c52c5f804f16982ef87d98b9.png"}, {"type": "image", "uri": "media/bdv2fail/f98fa713ba1b7dfdeb87df89.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0687/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 687, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the blue sieve to the right side of the cooker", "subtask": "move the blue sieve to the right side of the cooker", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aa1fa58b5174aaf7ac2c1f04.png"}, {"type": "image", "uri": "media/bdv2fail/dd165cde324e95956c3f9dce.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0688/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_20_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 688, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_20_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put eggplant beside but not on plate", "subtask": "put eggplant beside but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/07fac803b51973602f59c646.png"}, {"type": "image", "uri": "media/bdv2fail/8c14bccdab0c349ec99ce9b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0689/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 689, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the cloth to top from bottom", "subtask": "move the cloth to top from bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1af6cec5704137455abc579e.png"}, {"type": "image", "uri": "media/bdv2fail/c46bd3da739ec4bfb2c2403f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0690/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 690, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "fold the cloth from left to right", "subtask": "fold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e1d7db6b7e8df3da3102e87.png"}, {"type": "image", "uri": "media/bdv2fail/1ab3f91bd97396d840bc319f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0691/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_31_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 691, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_31_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the yellow towel left to right.", "subtask": "move the yellow towel left to right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a98c8a90fe2b4bba010eddaa.png"}, {"type": "image", "uri": "media/bdv2fail/d2a90ee89fa450e34bc4ea7e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0692/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_94_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 692, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_94_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "pick the yellow and white knife and put it on the right side of the hob", "subtask": "pick the yellow and white knife and put it on the right side of the hob", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bbb6dcd64df2e14cadebc265.png"}, {"type": "image", "uri": "media/bdv2fail/7e14d152bc20fb7e8286c955.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0693/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 693, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move orange cloth to the centre of the yellow toy carrot cone", "subtask": "move orange cloth to the centre of the yellow toy carrot cone", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8ca5d12b9780e495e5a7e395.png"}, {"type": "image", "uri": "media/bdv2fail/4415a60c56051ad8812520f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0694/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_11_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 694, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_11_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "unfold the cloth from left to right", "subtask": "unfold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b72d7abd114f073d69bc8d46.png"}, {"type": "image", "uri": "media/bdv2fail/c277e18386057f1a8ed725b0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0695/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_11_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 695, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_11_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "turn the cloth from top left to bottom right", "subtask": "turn the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f49a856219a4fabb2a242708.png"}, {"type": "image", "uri": "media/bdv2fail/50d7644d1676f5e85b858bff.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0697/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 697, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the red object at the bottom left of the left burner", "subtask": "put the red object at the bottom left of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14fd46ee99b5fa0573cfbed1.png"}, {"type": "image", "uri": "media/bdv2fail/97989ebe0b36cefdd42193b3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0698/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_71_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 698, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_71_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "unfold the cloth from right to left", "subtask": "unfold the cloth from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5440553c6781cee9501911ea.png"}, {"type": "image", "uri": "media/bdv2fail/c9f4b66ff468276d95077e97.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0699/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_30_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 699, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_30_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue fork to the upper left side of the stove", "subtask": "move the blue fork to the upper left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f63b382af30b50c6c65c850.png"}, {"type": "image", "uri": "media/bdv2fail/9b50f11dc64bcb706bffc9b2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0700/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_10_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 700, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_10_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the carrot from the left to the right and put it in top of the green cloth.", "subtask": "move the carrot from the left to the right and put it in top of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/433a332fedef5d025d4d3bce.png"}, {"type": "image", "uri": "media/bdv2fail/bf69525ae0fc9e844c647e73.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0701/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_28_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 701, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_28_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 249, "role": "outcome"}}
{"state": {"task": "move the metal strainer to the lower right of the table.", "subtask": "move the metal strainer to the lower right of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86753477bfea123aa804406e.png"}, {"type": "image", "uri": "media/bdv2fail/ec0b1f8275b802d6fdd3d2b5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0702/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 702, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the green object from the silver pot", "subtask": "remove the green object from the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f5009d123f10282ba5b8ac73.png"}, {"type": "image", "uri": "media/bdv2fail/a579474f3d1b86375cbf1b68.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0703/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_74_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 703, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_74_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the yellow corn toy between the burner and the sink.", "subtask": "move the yellow corn toy between the burner and the sink.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b08094972edd79ade776c1e8.png"}, {"type": "image", "uri": "media/bdv2fail/18d057d5ab58fd5e66790e2e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0704/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_59_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 704, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_59_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue spoon on the left side above the right burner", "subtask": "move the blue spoon on the left side above the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ddd85083e2e2dfeb9a04114e.png"}, {"type": "image", "uri": "media/bdv2fail/265a75f3a2f747e74f9bf53f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0705/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_41_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 705, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_41_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the knob to the left burner from the right burner.", "subtask": "move the knob to the left burner from the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c0542e1be9affae188b8156.png"}, {"type": "image", "uri": "media/bdv2fail/83a77635946d79fb0807eca6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0706/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_04_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 706, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_04_train", "episode_id": "20", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue cloth to the top of the pot.", "subtask": "move the blue cloth to the top of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d2787abaf5b3ee5d5292825.png"}, {"type": "image", "uri": "media/bdv2fail/4b0a158f06080203831bbff3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0707/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 707, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the red plastic fork on top of the green cloth.", "subtask": "move the red plastic fork on top of the green cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7089ec6ffd01b8ca61834720.png"}, {"type": "image", "uri": "media/bdv2fail/f7b5747946910c69979765a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0708/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 708, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the knob-like object to the right side of the table", "subtask": "move the knob-like object to the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ecdfd201f556737580f33240.png"}, {"type": "image", "uri": "media/bdv2fail/091ce238a2d40c0066dd6f26.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0709/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 709, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red toy scrubbing brush and place it on top of the orange cloth.", "subtask": "move the red toy scrubbing brush and place it on top of the orange cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7461e053a5b6e69086656776.png"}, {"type": "image", "uri": "media/bdv2fail/58274a51f9b000ffa42a7816.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0710/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 710, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "place the purple cloth on the right stove", "subtask": "place the purple cloth on the right stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c78b759ed6007c75311e9aed.png"}, {"type": "image", "uri": "media/bdv2fail/76d33557e52fa86270a0905b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0711/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_19_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 711, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_19_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the toy camera to the right side of the burner.", "subtask": "move the toy camera to the right side of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c42b354c174671d99573e52.png"}, {"type": "image", "uri": "media/bdv2fail/3092ccfbd3da8bc531aac88b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0712/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 712, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cheese into the pan.", "subtask": "move the cheese into the pan.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dda3c89983bad438cea10f4d.png"}, {"type": "image", "uri": "media/bdv2fail/413372a83226654b018cea8e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0713/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 713, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the blue fork at the bottom middle part of the burner", "subtask": "move the blue fork at the bottom middle part of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5034f4649dc30bbee8972565.png"}, {"type": "image", "uri": "media/bdv2fail/26b95e146e051b72dcd05fde.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0714/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_97_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 714, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_97_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the pot slightly left of the upper center of the stove.", "subtask": "move the pot slightly left of the upper center of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/916ba50708b0ed49534521e4.png"}, {"type": "image", "uri": "media/bdv2fail/f10172fb2c753a2ff2725b6f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0716/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_106_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 716, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_106_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move pan to left from right", "subtask": "move pan to left from right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4569e7869ae8d0624a3d1055.png"}, {"type": "image", "uri": "media/bdv2fail/1f434e508d902c274259396c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0717/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 717, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top left to bottom right", "subtask": "unfold the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f9aace0806cc7c58e7928e38.png"}, {"type": "image", "uri": "media/bdv2fail/f9aace0806cc7c58e7928e38.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0718/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_18_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 718, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_18_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the blue plastic toy from left to right", "subtask": "move the blue plastic toy from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1f434e508d902c274259396c.png"}, {"type": "image", "uri": "media/bdv2fail/4569e7869ae8d0624a3d1055.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0719/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 719, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the red object and place it on the left burner.", "subtask": "move the red object and place it on the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9d8611db6f5ba9488cf9e6b4.png"}, {"type": "image", "uri": "media/bdv2fail/630049a06895082a31d4f441.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0720/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 720, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue blob on the bottom to the opposite edge of the right burner.", "subtask": "move the blue blob on the bottom to the opposite edge of the right burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/61e0c06e9c0217d1efecf27f.png"}, {"type": "image", "uri": "media/bdv2fail/d36860e69368c7b98a43e5e1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0721/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_77_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 721, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_77_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "move the silver pot to the top left side of the table.", "subtask": "move the silver pot to the top left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ae970fe18df2b8bd6a9de3c4.png"}, {"type": "image", "uri": "media/bdv2fail/3216c5703c93ea25b751d76c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0722/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 722, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the corn in the silver pot", "subtask": "put the corn in the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4c90c19cef30f3bf468790a5.png"}, {"type": "image", "uri": "media/bdv2fail/f8ea5130969d36de31f2e4cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0723/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 723, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "unfold the cloth from left to right", "subtask": "unfold the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/31d598ccb4e581866f82a586.png"}, {"type": "image", "uri": "media/bdv2fail/da1aaa8f521f92dac00932b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0724/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_28_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 724, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_28_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the orange cloth", "subtask": "remove the silver pot from the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}, {"type": "image", "uri": "media/bdv2fail/0d5e34211222e84861c37f9d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0725/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 725, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the yellow spatula to the left top corner of the burner.", "subtask": "move the yellow spatula to the left top corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5921aeefd3e35236b3f68ab.png"}, {"type": "image", "uri": "media/bdv2fail/df7185d51294831a7bb0f6f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0726/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_55_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 726, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_55_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "take cucumber out of cup", "subtask": "take cucumber out of cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fb6b41b8cc088cd4cc6b102d.png"}, {"type": "image", "uri": "media/bdv2fail/83bc77017793b62ac52fb0e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0727/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_take_cucumber_out_of_cup_train/28", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 727, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_take_cucumber_out_of_cup_train", "episode_id": "28", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the brown and white stuffed animal to the left side of the table.", "subtask": "move the brown and white stuffed animal to the left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ad068cff729074332a509944.png"}, {"type": "image", "uri": "media/bdv2fail/16043f1b7d3b61bfab9dd8c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0728/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 728, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the red pepper pot on the bottom right corner of the burner.", "subtask": "move the red pepper pot on the bottom right corner of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b12600085f1b3fedb1f06c91.png"}, {"type": "image", "uri": "media/bdv2fail/c6b02a855e0038a796ef6352.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0729/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_78_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 729, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_78_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "remove spatula from plate sink", "subtask": "remove spatula from plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13039e8b6c9910c2454f22f6.png"}, {"type": "image", "uri": "media/bdv2fail/31513dd0ba13660b1c384c11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0730/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 730, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "put the yellow and orange spoon on the silver pot", "subtask": "put the yellow and orange spoon on the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0d01670dcbeb903c6f15302d.png"}, {"type": "image", "uri": "media/bdv2fail/00a7285242f1286e37a75331.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0731/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 731, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "take the pot and put it around but not on the green towel", "subtask": "take the pot and put it around but not on the green towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d3163e550be9737fab30421.png"}, {"type": "image", "uri": "media/bdv2fail/ae181b53b43ecae1217cc5e0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0732/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 732, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the pot to the right side of the table", "subtask": "move the pot to the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/609241498e32b42864514228.png"}, {"type": "image", "uri": "media/bdv2fail/760abfdd687db263eca229a3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0733/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 733, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the half orange in the half orange", "subtask": "put the half orange in the half orange", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/113282c5a87d7796e1031a58.png"}, {"type": "image", "uri": "media/bdv2fail/e6c67c205c9f478ef5da1c0c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0734/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 734, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "pick up violet allen key", "subtask": "pick up violet allen key", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b2e53e1d475fa13b0c08de7f.png"}, {"type": "image", "uri": "media/bdv2fail/f02f5f02edbb6edb7e694978.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0735/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_violet_Allen_key_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 735, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_violet_Allen_key_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put the banana to the top of the towel", "subtask": "put the banana to the top of the towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/737cfd8e9372b4ed6904e285.png"}, {"type": "image", "uri": "media/bdv2fail/e3af682a10fa11fdca0715d3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0736/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 736, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove the banana from the napkin from the right side of the table", "subtask": "remove the banana from the napkin from the right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f6334491f709ce9f247bad63.png"}, {"type": "image", "uri": "media/bdv2fail/91d129be15c2df2fd7b2cbe9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0737/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 737, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the silver pot between two burners", "subtask": "move the silver pot between two burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}, {"type": "image", "uri": "media/bdv2fail/1cb02cfb13d4475d5012d2f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0738/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 738, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the yellow cloth and place it in the lower left corner of the table", "subtask": "remove the silver pot from the yellow cloth and place it in the lower left corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/26b250a85408ec994349a45a.png"}, {"type": "image", "uri": "media/bdv2fail/a86bfea41967df9f35b88d61.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0739/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 739, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "move the pot and place it on the left burner.", "subtask": "move the pot and place it on the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a550b80e22d7a1620202c9a5.png"}, {"type": "image", "uri": "media/bdv2fail/80ff68f0b59e59fae064a89e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0740/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_102_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 740, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_102_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "place the blue spoon on the right side of the pot.", "subtask": "place the blue spoon on the right side of the pot.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/25f7eb5ce2919263dd6d6726.png"}, {"type": "image", "uri": "media/bdv2fail/a354e99bda4ac74498ef0d6a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0741/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 741, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the orange pot to the left side of the table", "subtask": "move the orange pot to the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/67168a679c95bbbd0e8b13c5.png"}, {"type": "image", "uri": "media/bdv2fail/c2a91645e1207fa80f9f69b4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0742/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_16_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 742, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_16_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the bottom right corner of the stove", "subtask": "move the blue spoon to the bottom right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13f46cd1486e593f53e980f3.png"}, {"type": "image", "uri": "media/bdv2fail/85a1b262c89b4616643254d1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0744/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 744, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the blue handle below the right burner", "subtask": "move the blue handle below the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/314bdf507e2340356d9737f0.png"}, {"type": "image", "uri": "media/bdv2fail/87f3040a964040a0fd38dc9b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0745/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_26_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 745, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_26_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the cloth from the middle to the top edge of the table.", "subtask": "move the cloth from the middle to the top edge of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f98fa713ba1b7dfdeb87df89.png"}, {"type": "image", "uri": "media/bdv2fail/c52c5f804f16982ef87d98b9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0746/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 746, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_00_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the blue silver spoon from right side of the burner to the center of the induction stove", "subtask": "move the blue silver spoon from right side of the burner to the center of the induction stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3cd21797217244e8c8f8aa71.png"}, {"type": "image", "uri": "media/bdv2fail/bd4a34150bbda26b93e0d803.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0747/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 747, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 249, "role": "outcome"}}
{"state": {"task": "move the silver pot on the left side of the burner", "subtask": "move the silver pot on the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5e934bf3cd6b4baa6be40f1.png"}, {"type": "image", "uri": "media/bdv2fail/bfa4813fda5f9fb4edcf1372.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0748/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 748, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth to the left side of the table", "subtask": "move the cloth to the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3cd12630ab5252b9769291f4.png"}, {"type": "image", "uri": "media/bdv2fail/218df4d9b5ed37da1cfbe488.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0749/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_01_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 749, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_01_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot lid from the pot to the yellow cloth on the top left of the table", "subtask": "move the silver pot lid from the pot to the yellow cloth on the top left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8ccfe7756ea7cb41987b5ec2.png"}, {"type": "image", "uri": "media/bdv2fail/9b31e86b8bb31773b6d23303.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0750/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 750, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "put away the fork from the green cloth", "subtask": "put away the fork from the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ef77ea3f1e543d0f6c8d156a.png"}, {"type": "image", "uri": "media/bdv2fail/ade9fc255f544a75f0ccb4a9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0751/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 751, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the blue fork to the upper left side of the stove", "subtask": "move the blue fork to the upper left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f63b382af30b50c6c65c850.png"}, {"type": "image", "uri": "media/bdv2fail/4f63b382af30b50c6c65c850.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0752/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_10_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 752, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_10_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the black burner to the top left", "subtask": "move the black burner to the top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d62a62d280800b3bdf618ebb.png"}, {"type": "image", "uri": "media/bdv2fail/36ce2c30811922a2172f1f7c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0753/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_04_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 753, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_04_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "pick up blue pen and put close by but not into drawer", "subtask": "pick up blue pen and put close by but not into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cfa56063aa926abebc96b070.png"}, {"type": "image", "uri": "media/bdv2fail/5971535d2a6cfb8965d04602.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0754/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 754, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth from left to right", "subtask": "move the cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/39950f2ce841b413543f9c9a.png"}, {"type": "image", "uri": "media/bdv2fail/986e630af04cfad7046679d8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0755/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 755, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the metal bowl to the upper right corner of the cooker.", "subtask": "move the metal bowl to the upper right corner of the cooker.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/45a2bb8bd12c8c54b1f02822.png"}, {"type": "image", "uri": "media/bdv2fail/9d4d5ffa08d28f97c0d6f272.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0757/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 757, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth from left to right of the table", "subtask": "move the cloth from left to right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/951bcbca10bfb008825032bd.png"}, {"type": "image", "uri": "media/bdv2fail/2d0e2c1c80a630c58f265efd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0758/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 758, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "remove the lid from the metal pot and put it on top of the green cloth", "subtask": "remove the lid from the metal pot and put it on top of the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/de9396e1269d42f4ff061a1c.png"}, {"type": "image", "uri": "media/bdv2fail/b7a77f5c541e9cbdd257349b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0759/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 759, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move the silver pot on the upper left corner of the stove", "subtask": "move the silver pot on the upper left corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5cadb056bed82d30b846ab5e.png"}, {"type": "image", "uri": "media/bdv2fail/9b2f2ee5f18e34fd469bbf64.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0760/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 760, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the green cloth to the right burner from the bottom left side of the stove.", "subtask": "move the green cloth to the right burner from the bottom left side of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0daa54e579b1052200f6164a.png"}, {"type": "image", "uri": "media/bdv2fail/bc4e76630abe1a643f2a2624.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0761/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_80_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 761, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_80_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "move the toy to the right side of the table.", "subtask": "move the toy to the right side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7a02bd12d15640b73901c947.png"}, {"type": "image", "uri": "media/bdv2fail/1c6648f50a3df4ce9c35e8c4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0762/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 762, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the can below the red ball", "subtask": "put the can below the red ball", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/228a07979502a009e2434732.png"}, {"type": "image", "uri": "media/bdv2fail/04750ea78f02d3b8f0cd3fab.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0763/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_04_train/26", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 763, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_04_train", "episode_id": "26", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put blueberries on plate sink", "subtask": "put blueberries on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e927d5f7aba53a029baead04.png"}, {"type": "image", "uri": "media/bdv2fail/d64f39ba48d655ea0af3e8df.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0764/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 764, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the toy mushroom to the right in front of the stove", "subtask": "move the toy mushroom to the right in front of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c83c88f550b82b62d6a1182f.png"}, {"type": "image", "uri": "media/bdv2fail/aa7c12fd9df6e4b0581bc6ea.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0765/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_32_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 765, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_32_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove cucumber from cup", "subtask": "remove cucumber from cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b3ca830dafcdf61fedee5ae.png"}, {"type": "image", "uri": "media/bdv2fail/7b3ca830dafcdf61fedee5ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0766/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 766, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "31", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "take and unfold the green cloth", "subtask": "take and unfold the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b74bcf5e5929ef70034964d0.png"}, {"type": "image", "uri": "media/bdv2fail/d0d86aa657d6283cffabcf19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0767/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_06_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 767, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_06_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move blue toy spoon over orange cloth on the right stove.", "subtask": "move blue toy spoon over orange cloth on the right stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/28ccd04f674bb1bce1a2ad4b.png"}, {"type": "image", "uri": "media/bdv2fail/471f97d879ed91e552cddd0d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0768/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 768, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "take the white spoon from the silver pot and put it on the lower left corner of the hob.", "subtask": "take the white spoon from the silver pot and put it on the lower left corner of the hob.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d37ddaed5284e4f411886492.png"}, {"type": "image", "uri": "media/bdv2fail/5b52a77d151af3108c19293d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0769/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_22_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 769, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_22_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "unfold the cloth from bottom left to top right", "subtask": "unfold the cloth from bottom left to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eac307b506d783c17fd59a54.png"}, {"type": "image", "uri": "media/bdv2fail/497d4432fae10611f25c2d96.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0770/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 770, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_25_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the blue plastic toy to left from right", "subtask": "move the blue plastic toy to left from right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4569e7869ae8d0624a3d1055.png"}, {"type": "image", "uri": "media/bdv2fail/1f434e508d902c274259396c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0771/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 771, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "place the pot in front of the left burner", "subtask": "place the pot in front of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70c07fab385c391fe9358c57.png"}, {"type": "image", "uri": "media/bdv2fail/5fa7b8078edcbdb117d2ebcd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0772/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_33_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 772, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_33_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "take the silver lid off the silver pot", "subtask": "take the silver lid off the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aacd5e9e67e82854bf9fcc71.png"}, {"type": "image", "uri": "media/bdv2fail/c3e1a84c5b54d1062518eb5d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0773/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 773, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "open the pot", "subtask": "open the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e7fe93c1274cfade03a55d11.png"}, {"type": "image", "uri": "media/bdv2fail/9295ce60e99485d7d9b96276.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0774/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 774, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "move the drumstick to the top left edge of the cooker", "subtask": "move the drumstick to the top left edge of the cooker", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/af6a6b677d753e1a91874be6.png"}, {"type": "image", "uri": "media/bdv2fail/02f45d49d0d0dea2f90cd6d5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0775/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 775, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the blue plastic spoon to the bottom of the left burner", "subtask": "move the blue plastic spoon to the bottom of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1383ffb9ad397d8e247dbc42.png"}, {"type": "image", "uri": "media/bdv2fail/812c35b0fe4614f3cbcab8ea.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0776/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_02_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 776, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_02_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the purple cloth next to the bowl next to the sushi.", "subtask": "move the purple cloth next to the bowl next to the sushi.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6dab21dc75a998985940193.png"}, {"type": "image", "uri": "media/bdv2fail/83f13841d65e1577a3943e40.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0777/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_17_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 777, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_17_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the silver pot on the lower left corner of the burner", "subtask": "move the silver pot on the lower left corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3f6f9f39bb25ea56d7c2dd7d.png"}, {"type": "image", "uri": "media/bdv2fail/87fe80173b1d745435613f00.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0778/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 778, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "unfold the cloth from middle to right", "subtask": "unfold the cloth from middle to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/647a91c634ef5e7e06815d78.png"}, {"type": "image", "uri": "media/bdv2fail/647a91c634ef5e7e06815d78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0779/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_12_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 779, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_12_train", "episode_id": "5", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upper right side of the stove", "subtask": "move the silver pot to the upper right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/117646b5fdd635f4290a7309.png"}, {"type": "image", "uri": "media/bdv2fail/6ba0259d82f991cf797de1ed.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0780/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 780, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "pick up blue pen and put into drawer", "subtask": "pick up blue pen and put into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4233b866de77b685595cea8.png"}, {"type": "image", "uri": "media/bdv2fail/dbb82d5ec344c72ef5bdee42.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0781/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 781, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "turn faucet front to right", "subtask": "turn faucet front to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/997d63f6b7c05e7489de45d3.png"}, {"type": "image", "uri": "media/bdv2fail/ac936a6477def5e8a67ad9ca.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0782/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 782, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e019d31f930011c354deed83.png"}, {"type": "image", "uri": "media/bdv2fail/7ae1526e1dfc7f8992aab652.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0783/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 783, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_15_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "turn faucet front to right", "subtask": "turn faucet front to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d8873cb7ad14a3d00fe87060.png"}, {"type": "image", "uri": "media/bdv2fail/bcee274e4f655713203d1adc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0784/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 784, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_turn_faucet_front_to_right_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top right to bottom left", "subtask": "unfold the cloth from top right to bottom left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/432f4998f546e07ad2157919.png"}, {"type": "image", "uri": "media/bdv2fail/732b67cb45181ea6a66044c2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0785/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 785, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom left to top right", "subtask": "fold the cloth from bottom left to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13fd1cbb0aff0d0e55e4d199.png"}, {"type": "image", "uri": "media/bdv2fail/5b744a0691aea2686b5c76fa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0786/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_23_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 786, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_23_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove the silver pot from the left burner", "subtask": "remove the silver pot from the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5c7ec7b5cf3b17b56984ef38.png"}, {"type": "image", "uri": "media/bdv2fail/08f54d15458c03e556da39fa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0787/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 787, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "take the silver pot to the top of the left burner", "subtask": "take the silver pot to the top of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/84a7a0fa0c9d8193ab4633a9.png"}, {"type": "image", "uri": "media/bdv2fail/bcad9b238f6ede573da8bc0a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0788/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_16_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 788, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_16_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot to the top of the blue curtains", "subtask": "move the silver pot to the top of the blue curtains", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91019a6c95a080684ca23b8b.png"}, {"type": "image", "uri": "media/bdv2fail/ab07e2b7eb87cd9daef52745.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0789/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 789, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put corn in bowl sink", "subtask": "put corn in bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/63ae23d93612a7ee2a93e2d3.png"}, {"type": "image", "uri": "media/bdv2fail/bdf054964393383338b02ae6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0790/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/43", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 790, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "43", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "pick up scissors and put close to but not inside drawer", "subtask": "pick up scissors and put close to but not inside drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e965f9d8e06b966ff135b275.png"}, {"type": "image", "uri": "media/bdv2fail/2ffe84b552b721d6587847fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0791/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train/38", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 791, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train", "episode_id": "38", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the banana to the pot", "subtask": "move the banana to the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1d4b2d6926ad1d44ed72cdf4.png"}, {"type": "image", "uri": "media/bdv2fail/e21151e72f00c5074b9183d5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0792/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 792, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "take the yellow duck out of the drawer and put it on the left side of the table", "subtask": "take the yellow duck out of the drawer and put it on the left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/963bb78ca6ace4a99b975dc0.png"}, {"type": "image", "uri": "media/bdv2fail/d226cbcc46441d87d3559117.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0793/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_18_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 793, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_18_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 249, "role": "outcome"}}
{"state": {"task": "move the silver pot lid to the pot from the yellow cloth on the top right of the table", "subtask": "move the silver pot lid to the pot from the yellow cloth on the top right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9b31e86b8bb31773b6d23303.png"}, {"type": "image", "uri": "media/bdv2fail/8ccfe7756ea7cb41987b5ec2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0794/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 794, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "0", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 251, "role": "outcome"}}
{"state": {"task": "take the orange towel and move it to the left side of the burner.", "subtask": "take the orange towel and move it to the left side of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d6b6fc758392eb63c0c3f2fc.png"}, {"type": "image", "uri": "media/bdv2fail/8d48d02d3e4aacf580c2f92d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0795/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 795, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "picking the blue plastic item from the stove and place it to the left burner on the left side.", "subtask": "picking the blue plastic item from the stove and place it to the left burner on the left side.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a3c3168a4bc7d69cb2264d99.png"}, {"type": "image", "uri": "media/bdv2fail/f287190cff6cff02b0cbedde.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0796/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 796, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 253, "role": "outcome"}}
{"state": {"task": "move cloth to the bottom right", "subtask": "move cloth to the bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/83e0ec47ae798338e36124f9.png"}, {"type": "image", "uri": "media/bdv2fail/7714c444e6c35953a15fd3d5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0797/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_08_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 797, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_08_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom right side of the stove", "subtask": "move the silver pot to the bottom right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/34cd516bb80990b2213577f0.png"}, {"type": "image", "uri": "media/bdv2fail/9d66e437eff4233190b73d4c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0798/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 798, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the cloth to left from right", "subtask": "move the cloth to left from right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9fd3b47bcf9d897c052e765e.png"}, {"type": "image", "uri": "media/bdv2fail/3d5da668f57dbae50dbca495.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0799/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 799, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the spatula to the bottom right corner of the stove", "subtask": "put the spatula to the bottom right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/558a45de0e6cfd97bb8317ba.png"}, {"type": "image", "uri": "media/bdv2fail/104eabd0e01ffbb31032e7fa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0800/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_00_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 800, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_00_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the hot dog on the right side of the blue spoon", "subtask": "put the hot dog on the right side of the blue spoon", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/36d55bca860aecd69a086080.png"}, {"type": "image", "uri": "media/bdv2fail/5fcdc4abff62660b2c2460e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0801/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 801, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold the blue cloth from bottom left to top right", "subtask": "fold the blue cloth from bottom left to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a45c5c95ea48dfbcd86470ff.png"}, {"type": "image", "uri": "media/bdv2fail/bd11a64c29962634104ac257.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0802/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_06_train/28", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 802, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_06_train", "episode_id": "28", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the gray pot on the lower left side.", "subtask": "move the gray pot on the lower left side.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d5a5c43f38c89a29e6a15a69.png"}, {"type": "image", "uri": "media/bdv2fail/ca53019445c5da9fb337f630.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0804/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_88_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 804, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_88_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "take the white spoon and put it between the center of the burner", "subtask": "take the white spoon and put it between the center of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bcdf71b4c26de9e7e26f0c65.png"}, {"type": "image", "uri": "media/bdv2fail/8552aed0e825f00fd9a77869.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0805/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 805, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the blue towel to the right side of the burner", "subtask": "move the blue towel to the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9a6418fb9aada573bf773f2.png"}, {"type": "image", "uri": "media/bdv2fail/bfd5bcd54254f2b401a88dbf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0806/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 806, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the red strawberry on the left burner", "subtask": "put the red strawberry on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/73ee22b287f8cbcb4fdde66f.png"}, {"type": "image", "uri": "media/bdv2fail/65d0980984fb024e62525f41.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0807/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_72_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 807, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_72_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "moves the green star-shaped object to top.", "subtask": "moves the green star-shaped object to top.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fb1f7a299fc1a5a0c850d91f.png"}, {"type": "image", "uri": "media/bdv2fail/6e09576c6a4b866c8aed8fda.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0808/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 808, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the silver pot to the upper left corner of the table", "subtask": "move the silver pot to the upper left corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ccf20ad56b59058f6bb0e28d.png"}, {"type": "image", "uri": "media/bdv2fail/6e9363e4aaddc8074827b76d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0809/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 809, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold cloth from top left to bottom right", "subtask": "fold cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7895da3647a6d617a0a34853.png"}, {"type": "image", "uri": "media/bdv2fail/7895da3647a6d617a0a34853.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0810/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train/33", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 810, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_10_train", "episode_id": "33", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the big spoon to the right of the silver pan", "subtask": "put the big spoon to the right of the silver pan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bb0e95a71adc53331869e83a.png"}, {"type": "image", "uri": "media/bdv2fail/e2b4597bee4ea775dfb17005.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0811/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_06_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 811, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_06_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the green toy underneath of purple cloth", "subtask": "put the green toy underneath of purple cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4fabef0ffb84d546246736c7.png"}, {"type": "image", "uri": "media/bdv2fail/1f2210baa147340cd6314b54.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0812/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 812, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the yellow spatula to top right of burner", "subtask": "move the yellow spatula to top right of burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c67cca250de9fd7b64d2f5dd.png"}, {"type": "image", "uri": "media/bdv2fail/a307de57467c1d3ddf3a6255.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0813/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_08_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 813, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_08_train", "episode_id": "3", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the can of fake food to the top right of the stove.", "subtask": "move the can of fake food to the top right of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e63a1c16f1c917b268df1620.png"}, {"type": "image", "uri": "media/bdv2fail/923ff77afbfae10648ac9d69.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0814/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 814, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_12_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "take the lion toy and place it above the drawer", "subtask": "take the lion toy and place it above the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/acd0111bd5e8882b17aef9c0.png"}, {"type": "image", "uri": "media/bdv2fail/df99b5cd92ae3cabefe1b28a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0815/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/41", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 815, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "41", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put cup to counter from sink", "subtask": "put cup to counter from sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6863c54f5074103a8182f723.png"}, {"type": "image", "uri": "media/bdv2fail/618f735c173259d70592d0ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0816/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/43", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 816, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "43", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "fold the cloth from the left to the right", "subtask": "fold the cloth from the left to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/551eb149f3976e4ed2af0701.png"}, {"type": "image", "uri": "media/bdv2fail/551eb149f3976e4ed2af0701.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0817/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_18_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 817, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_18_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the eggplant broccoli in the drawer", "subtask": "put the eggplant broccoli in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b7c699726c0764aa80b18728.png"}, {"type": "image", "uri": "media/bdv2fail/baaae7251684e6571a4a277c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0818/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_05_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 818, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_05_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the cloth to the left from the bottom right of the table", "subtask": "move the cloth to the left from the bottom right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c265207fe63c9bc0271de7fc.png"}, {"type": "image", "uri": "media/bdv2fail/7629502823b7866901cf5b1c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0819/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_22_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 819, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_22_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the silver pot to the top of the green cloth", "subtask": "move the silver pot to the top of the green cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8339c0e465a9ce5e768cde06.png"}, {"type": "image", "uri": "media/bdv2fail/058e5b0114c6c0797faa1b07.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0820/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 820, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_24_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "touch the banana in the drawer", "subtask": "touch the banana in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cf252df7b29567eafc88f8af.png"}, {"type": "image", "uri": "media/bdv2fail/c7311c62928a6c661f31a8b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0821/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 821, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the yellow cloth on the right side of the stove", "subtask": "move the yellow cloth on the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/707b283431514669b57c550a.png"}, {"type": "image", "uri": "media/bdv2fail/d420a725df081179367983fa.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0822/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_21_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 822, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_21_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue cloth to the bottom right of the burner.", "subtask": "move the blue cloth to the bottom right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/417be5b92ba7ce76de67bba6.png"}, {"type": "image", "uri": "media/bdv2fail/4e6faddc9d115edea4ae61fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0823/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_103_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 823, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_103_train", "episode_id": "12", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "moving the toy ball from the center bottom of two burner to center top.", "subtask": "moving the toy ball from the center bottom of two burner to center top.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/78912662b490240b07844e31.png"}, {"type": "image", "uri": "media/bdv2fail/ce5918e237f8e51bae7fbb1f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0824/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 824, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "remove the red spoon from the yellow cloth", "subtask": "remove the red spoon from the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d62a13dca5f3b8c3811ba8d8.png"}, {"type": "image", "uri": "media/bdv2fail/b9b90dfd3c0b69edbcb498f2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0826/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 826, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_31_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "take the yellow banana and put beside but not in the left bottom of the table", "subtask": "take the yellow banana and put beside but not in the left bottom of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a61f3f332984fa96881c32be.png"}, {"type": "image", "uri": "media/bdv2fail/fc9002a13b7ca9b2e6f7a693.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0827/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/26", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 827, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "26", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "put the yellow object at the bottom left side of the table", "subtask": "put the yellow object at the bottom left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/10de0d2a4995092f0246f042.png"}, {"type": "image", "uri": "media/bdv2fail/9838139ad91a1089b5c8264b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0828/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 828, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train", "episode_id": "31", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the silver pot on the top left side of the table.", "subtask": "put the silver pot on the top left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/995aff5f8aec2739c6e6659e.png"}, {"type": "image", "uri": "media/bdv2fail/ec81563932a7af2df0c4108e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0829/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 829, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the towel to the left side of the stove", "subtask": "move the towel to the left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/efc707c5addceffe4ea88030.png"}, {"type": "image", "uri": "media/bdv2fail/5fcabd0ef6917aa6576fcd21.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0830/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_11_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 830, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_11_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the yellow spoon to the bottom of the right burner and blue cloth.", "subtask": "move the yellow spoon to the bottom of the right burner and blue cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6209230ade8e2a33b4370109.png"}, {"type": "image", "uri": "media/bdv2fail/36ced3fe78a1599472814256.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0831/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_69_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 831, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_69_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "take blueberries off plate sink", "subtask": "take blueberries off plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/455aa887caf2b45b72a154e1.png"}, {"type": "image", "uri": "media/bdv2fail/2eb11663346f91df03804913.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0832/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_blueberries_off_plate_sink_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 832, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_blueberries_off_plate_sink_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "cover the pot", "subtask": "cover the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4fba0a9455049f1be4923ad8.png"}, {"type": "image", "uri": "media/bdv2fail/f2f20c1afa362c069a4dafdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0833/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 833, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "move the white and orange object around but not on the right burner", "subtask": "move the white and orange object around but not on the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d2041f4c15a4c9033cbdba7d.png"}, {"type": "image", "uri": "media/bdv2fail/fa47e5cd854727c48503f5fd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0834/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 834, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "put the yellow cloth between the burners of the stove.", "subtask": "put the yellow cloth between the burners of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7d1c5b83c7a37ddf983bc514.png"}, {"type": "image", "uri": "media/bdv2fail/9e6a9797eaae8eff5d455af6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0835/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 835, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move silver pot to back left corner of stove top", "subtask": "move silver pot to back left corner of stove top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc5c86687fd9524e7c522f98.png"}, {"type": "image", "uri": "media/bdv2fail/e68ad6b23e94ea0074e1d640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0836/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_103_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 836, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_103_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the red spoon in the right burner", "subtask": "put the red spoon in the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a31a96a803cb2c74b124773.png"}, {"type": "image", "uri": "media/bdv2fail/3a31a96a803cb2c74b124773.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0837/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 837, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the silver pot to the top right side of the table", "subtask": "move the silver pot to the top right side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ac07bb9fabdcee6bc274b0f.png"}, {"type": "image", "uri": "media/bdv2fail/3cf39080b46015547b130b17.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0838/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 838, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_08_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold the cloth from right to left", "subtask": "fold the cloth from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/847064113287e48a3fdf2aea.png"}, {"type": "image", "uri": "media/bdv2fail/9254f605328ff9ed2544ff6f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0839/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 839, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom left to top right", "subtask": "fold the cloth from bottom left to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f2ae95f5cb6359b37f6ef00.png"}, {"type": "image", "uri": "media/bdv2fail/a8056f8e34164318e41b37e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0840/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_22_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 840, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_22_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the object in the middle of the burner switch", "subtask": "move the object in the middle of the burner switch", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a70942ceb10ed51e4e98191.png"}, {"type": "image", "uri": "media/bdv2fail/222f9056da8b6f215b8cb568.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0841/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 841, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the yellow knife on the orange cloth.", "subtask": "put the yellow knife on the orange cloth.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0d5e34211222e84861c37f9d.png"}, {"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0842/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 842, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "unfold the cloth from right to left side", "subtask": "unfold the cloth from right to left side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d207839ff359e7890cf0c7da.png"}, {"type": "image", "uri": "media/bdv2fail/d207839ff359e7890cf0c7da.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0843/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 843, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "7", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the hot dog from the left side of the blue spoon", "subtask": "remove the hot dog from the left side of the blue spoon", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5fcdc4abff62660b2c2460e8.png"}, {"type": "image", "uri": "media/bdv2fail/36d55bca860aecd69a086080.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0844/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 844, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the yellow cloth in the right side of the silver pot", "subtask": "move the yellow cloth in the right side of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b91bfd5c180545c3532b835f.png"}, {"type": "image", "uri": "media/bdv2fail/03362fe902c4e5de5b1c60d2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0845/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 845, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put cup around but not on plate", "subtask": "put cup around but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9eba0e1cf2b8fca02b17a05a.png"}, {"type": "image", "uri": "media/bdv2fail/05e174d4bfb7f2c026e589f6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0846/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/30", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 846, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "30", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the right of the green spatula", "subtask": "move the yellow cloth to the right of the green spatula", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9b192242bbf45523a3d87730.png"}, {"type": "image", "uri": "media/bdv2fail/4cc99c0006c06bb04fdb2f9e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0847/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 847, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the green towel to upper left of the sink.", "subtask": "move the green towel to upper left of the sink.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/36945f57034a1427eb12cf0c.png"}, {"type": "image", "uri": "media/bdv2fail/8556f54b205885711dd44d21.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0848/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 848, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_04_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the yellow knife to the top of the can.", "subtask": "move the yellow knife to the top of the can.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ff76cb4da0d554edd36bd7c8.png"}, {"type": "image", "uri": "media/bdv2fail/bf53c4f8f687a9663acab521.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0849/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 849, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train", "episode_id": "20", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put the blue stuffed toy above the drawer", "subtask": "put the blue stuffed toy above the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8b9e16d4306697fcc8b37abf.png"}, {"type": "image", "uri": "media/bdv2fail/d3ee15198715f4a4f7465693.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0850/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 850, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "31", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "take the green broccoli off the silver pot", "subtask": "take the green broccoli off the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aacd5e9e67e82854bf9fcc71.png"}, {"type": "image", "uri": "media/bdv2fail/c3e1a84c5b54d1062518eb5d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0851/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 851, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "9", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the small box on the cloth", "subtask": "remove the small box on the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7a9bd5faa1c4a66528220d4b.png"}, {"type": "image", "uri": "media/bdv2fail/f86c40c780f7aff483e8b659.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0853/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_16_train/33", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 853, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_16_train", "episode_id": "33", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move silver pot onto the left burner.", "subtask": "move silver pot onto the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2fc973adb787cd5ebf1b48dd.png"}, {"type": "image", "uri": "media/bdv2fail/14370880da5faf7909bfac78.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0854/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 854, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue object to the upper left side of the right burner", "subtask": "move the blue object to the upper left side of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/946b113097009abfd7b9a8b2.png"}, {"type": "image", "uri": "media/bdv2fail/0bcaf4d1b95d74c883342cb3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0855/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_70_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 855, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_70_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "fold the cloth from the bottom left to the top right", "subtask": "fold the cloth from the bottom left to the top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/54308b868395e93a1eebf096.png"}, {"type": "image", "uri": "media/bdv2fail/34fa19db34d7037d3bbdb71e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0856/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train/37", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 856, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_03_train", "episode_id": "37", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot to the left top burner.", "subtask": "move the silver pot to the left top burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/51aa2ba54aa20671d6b510fe.png"}, {"type": "image", "uri": "media/bdv2fail/11dcf36cec255fe24d4be854.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0857/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_00_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 857, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_00_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the brown lid on the brown pot", "subtask": "put the brown lid on the brown pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e15ddef653eba328a364d00.png"}, {"type": "image", "uri": "media/bdv2fail/74934245d7b66974af3ff857.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0858/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 858, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "make cloth stand on left side of the table", "subtask": "make cloth stand on left side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6111dcdff6aa9e0625fba0b5.png"}, {"type": "image", "uri": "media/bdv2fail/55b033b14db6487eae16558a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0859/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_16_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 859, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_16_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "take corn out of bowl sink", "subtask": "take corn out of bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/765ddaafd48c80831a186e00.png"}, {"type": "image", "uri": "media/bdv2fail/97ae190e3c9e1d4bf8f436b0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0860/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 860, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "31", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "place the green object inside the red toy box", "subtask": "place the green object inside the red toy box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e91d0dd17e8e5cae5243902a.png"}, {"type": "image", "uri": "media/bdv2fail/6751c9bd7fa2124cc4ebd2fc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0861/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_01_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 861, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_01_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the silver pot on the upper let side of the left burner", "subtask": "put the silver pot on the upper let side of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0e5d31ab414a9355f8035322.png"}, {"type": "image", "uri": "media/bdv2fail/76ea18347bbc7a6344889203.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0862/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_42_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 862, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_42_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "put the red object on the left side of the table.", "subtask": "put the red object on the left side of the table.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/68cd038d94edc6d07fb40bd4.png"}, {"type": "image", "uri": "media/bdv2fail/109070970ba60ad8fae0729f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0863/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 863, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "take the red knob to the bottom right of the burner.", "subtask": "take the red knob to the bottom right of the burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/53789d6bb02a991fdcdeaad5.png"}, {"type": "image", "uri": "media/bdv2fail/8560fe0afeb1efd46483c92d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0864/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 864, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the blue napkin to bottom from top of the right side induction stove burner", "subtask": "move the blue napkin to bottom from top of the right side induction stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf3d01b1f3bb0efa6df18d65.png"}, {"type": "image", "uri": "media/bdv2fail/6ece60bfe416b3f626ca9cac.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0865/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_62_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 865, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_62_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "pick up screwdriver and put into drawer", "subtask": "pick up screwdriver and put into drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6a8614ceb4fc31e2b2a784ef.png"}, {"type": "image", "uri": "media/bdv2fail/2474d52ded7ccd4d94c80342.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0866/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_glue_and_put_into_drawer_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 866, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_glue_and_put_into_drawer_train", "episode_id": "16", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the lid on the silver pot in the left side of table", "subtask": "put the lid on the silver pot in the left side of table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/30a60024ae33916c2719dec6.png"}, {"type": "image", "uri": "media/bdv2fail/0b72947d31006188ca2038c0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0868/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 868, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "unfold cloth from top to bottom", "subtask": "unfold cloth from top to bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/608dc9da736222bdc712e002.png"}, {"type": "image", "uri": "media/bdv2fail/33105808f536350243bb6738.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0869/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 869, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc5ec89012aeb7227cfd52ae.png"}, {"type": "image", "uri": "media/bdv2fail/cc5ec89012aeb7227cfd52ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0871/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 871, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "put corn next to but not in bowl sink", "subtask": "put corn next to but not in bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/63ae23d93612a7ee2a93e2d3.png"}, {"type": "image", "uri": "media/bdv2fail/bdf054964393383338b02ae6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0872/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/43", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 872, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "43", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "moves the ribbon from right of the cloth to top", "subtask": "moves the ribbon from right of the cloth to top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/43eea7b49ab6e0a76025341e.png"}, {"type": "image", "uri": "media/bdv2fail/4ebbc64e9b733f0257e30a7e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0873/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/34", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 873, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "34", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot above the left burner", "subtask": "move the silver pot above the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f17bd502ec2098873c8e719.png"}, {"type": "image", "uri": "media/bdv2fail/030b33fdc020da1bad202c68.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0874/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 874, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top left to to bottom right", "subtask": "unfold the cloth from top left to to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7ecf26b9e516ed6d1d890d3f.png"}, {"type": "image", "uri": "media/bdv2fail/f515bd48d6bc5d5e0a608e76.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0875/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_13_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 875, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_13_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the pear to the bottom left of the burner", "subtask": "move the pear to the bottom left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/31e659ab4e8709fe59fd0106.png"}, {"type": "image", "uri": "media/bdv2fail/f291024a02fa39a39920610e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0876/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 876, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put cup to counter from sink", "subtask": "put cup to counter from sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e871c8d4ae3e86b344448ec.png"}, {"type": "image", "uri": "media/bdv2fail/8e871c8d4ae3e86b344448ec.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0877/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/26", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 877, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "26", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the pan somewhat away from the orange cloth", "subtask": "move the pan somewhat away from the orange cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/75a288e8ac9cdc8c8095a645.png"}, {"type": "image", "uri": "media/bdv2fail/213c85e31f627f95ae06b697.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0878/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_00_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 878, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_00_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put cup on plate", "subtask": "put cup on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d70567d9bf2c993a715fe709.png"}, {"type": "image", "uri": "media/bdv2fail/3677a790ecb1ad38d831c2e2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0879/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/24", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 879, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "24", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "remove the blue spoon from the right side of the burner", "subtask": "remove the blue spoon from the right side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3892f532518a48b8a2a86cdc.png"}, {"type": "image", "uri": "media/bdv2fail/3892f532518a48b8a2a86cdc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0880/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 880, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "9", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the stove to the right side of the yellow cloth", "subtask": "move the stove to the right side of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ce2ab4922adfc005e76b76a3.png"}, {"type": "image", "uri": "media/bdv2fail/89c7a1c81e063b4eef1f0774.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0881/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_03_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 881, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_03_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the metal bowl with the robot and place it on the far edge of the stove.", "subtask": "move the metal bowl with the robot and place it on the far edge of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13beef7979ad5e47dac9fd39.png"}, {"type": "image", "uri": "media/bdv2fail/f30ec8d748400e79b73fc16a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0882/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_91_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 882, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_91_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "move a blue cloth to the bottom right side of the toy oven.", "subtask": "move a blue cloth to the bottom right side of the toy oven.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a73fbf24b4cd93e1bf761400.png"}, {"type": "image", "uri": "media/bdv2fail/dbafda0d494d99390f83bc12.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0883/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_03_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 883, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_03_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "take out the chocolate above the drawer and put it on the bottom center of the table", "subtask": "take out the chocolate above the drawer and put it on the bottom center of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c91ad779f194f082b891595c.png"}, {"type": "image", "uri": "media/bdv2fail/f9d59305309decdffbf7ebfb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0884/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_12_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 884, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_12_train", "episode_id": "31", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "move the silver pot on the lower right side", "subtask": "move the silver pot on the lower right side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ff7e7cc22766fcc3ff5fabd.png"}, {"type": "image", "uri": "media/bdv2fail/0463f22e3e28f84cc5da8021.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0886/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_87_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 886, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_87_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the purple cloth to the left side of the blue glass", "subtask": "move the purple cloth to the left side of the blue glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4ff19b140bb24688f4a1b7f1.png"}, {"type": "image", "uri": "media/bdv2fail/854e9420634017ba7855159c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0887/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 887, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the chicken leg in the silver pot", "subtask": "put the chicken leg in the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2f1425179ff6ab0dce75cc29.png"}, {"type": "image", "uri": "media/bdv2fail/605c2698db40b23da2e36801.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0888/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 888, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "grab the yellow square to the middle and moved it from the bottom left edge", "subtask": "grab the yellow square to the middle and moved it from the bottom left edge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/66b4d43bcb2a9242a54413f4.png"}, {"type": "image", "uri": "media/bdv2fail/44ce6fdadd8eeb0c8a1afdf9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0889/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 889, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "put blueberries close to but not on plate sink", "subtask": "put blueberries close to but not on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e927d5f7aba53a029baead04.png"}, {"type": "image", "uri": "media/bdv2fail/d64f39ba48d655ea0af3e8df.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0890/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 890, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the duck from right to left", "subtask": "move the duck from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f3e3872f7b27130175efb216.png"}, {"type": "image", "uri": "media/bdv2fail/f3e3872f7b27130175efb216.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0891/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/21", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 891, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "21", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the cloth from the top left of the table to the bottom right.", "subtask": "move the cloth from the top left of the table to the bottom right.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3c00ab61e417dfcc122bd2fc.png"}, {"type": "image", "uri": "media/bdv2fail/f92c2170a61483ff8b412b20.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0892/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 892, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the purple towel to the top left corner of the burner", "subtask": "move the purple towel to the top left corner of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f5134749a746fc067631c6c.png"}, {"type": "image", "uri": "media/bdv2fail/c7aa48d80f7c2ff57405164d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0893/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_60_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 893, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_60_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put corn in bowl sink", "subtask": "put corn in bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a59d7176f4cfde0275bc3597.png"}, {"type": "image", "uri": "media/bdv2fail/3124b5ddd6d73d5c63413db7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0894/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 894, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put the silver pot on top of the silver spoon.", "subtask": "put the silver pot on top of the silver spoon.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0592cb5c5172e9ecc9d45806.png"}, {"type": "image", "uri": "media/bdv2fail/2e631394f8d985105fa1f975.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0895/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 895, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put cup near but not on plate", "subtask": "put cup near but not on plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf27f77209ad7ef28e7779b0.png"}, {"type": "image", "uri": "media/bdv2fail/cf119044eef1f2d1c1f0beaf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0897/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/39", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 897, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "39", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "take spoon out of bowl sink", "subtask": "take spoon out of bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b3427730f8cd30d7fce6a966.png"}, {"type": "image", "uri": "media/bdv2fail/e06dbcaff65ebf0d14470750.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0898/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/33", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 898, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "33", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the blue spoon to the bottom of the yellow cloth", "subtask": "move the blue spoon to the bottom of the yellow cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ca35ef55aa8fb8d1d24f907c.png"}, {"type": "image", "uri": "media/bdv2fail/340f1b77d05d85555e644f2b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0899/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/30", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 899, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "30", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "close microwave", "subtask": "close microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd143b1aff093837e28a456b.png"}, {"type": "image", "uri": "media/bdv2fail/3b1013bd49180562d6f01627.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0900/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_open_microwave_train/44", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 900, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_open_microwave_train", "episode_id": "44", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 217, "role": "outcome"}}
{"state": {"task": "move the can to the top of the left burner", "subtask": "move the can to the top of the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d8fe49c3e772f22e6f6be79c.png"}, {"type": "image", "uri": "media/bdv2fail/d45104f99887632d068fbdbb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0901/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 901, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "put cucumber close by but not in cup", "subtask": "put cucumber close by but not in cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dbce06c934d3d93ffac65ae1.png"}, {"type": "image", "uri": "media/bdv2fail/23b9766f583bf6909b474dc1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0902/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/34", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 902, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "34", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the purple figure at the top left of the wooden box", "subtask": "place the purple figure at the top left of the wooden box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/07af6127d9cb20e9964817d8.png"}, {"type": "image", "uri": "media/bdv2fail/dde9e92b963f81b80104fd8f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0903/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_12_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 903, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_12_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "remove the orange fish from the silver bowl", "subtask": "remove the orange fish from the silver bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91da124d73ab01a9507065c6.png"}, {"type": "image", "uri": "media/bdv2fail/91da124d73ab01a9507065c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0904/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 904, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "15", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the can near the small orange bowl.", "subtask": "move the can near the small orange bowl.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6d0dda7dc49da7a5ba4f0398.png"}, {"type": "image", "uri": "media/bdv2fail/bd48e75c8316045c8bf86c53.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0905/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_16_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 905, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_16_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the orange object to the bottom side of the table", "subtask": "move the orange object to the bottom side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/33f3ef3037f795a6ccc9838e.png"}, {"type": "image", "uri": "media/bdv2fail/3e7bea1f8d7ab41dbc267784.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0906/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 906, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the upper left of the stove", "subtask": "move the yellow cloth to the upper left of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/87883a8c33ae5d84b5115a16.png"}, {"type": "image", "uri": "media/bdv2fail/1cfcc3d8a351f7dffa3a4055.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0907/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 907, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth from right to left", "subtask": "move the cloth from right to left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/97de6df2eb6720e14986adb1.png"}, {"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0908/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 908, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the cloth from left corner of the table to the top of the table", "subtask": "move the cloth from left corner of the table to the top of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/080af74866c974dbf4e2604f.png"}, {"type": "image", "uri": "media/bdv2fail/2a0e138e25f7a006347ac757.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0910/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 910, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_09_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "place the blue brush on top of the purple cloth", "subtask": "place the blue brush on top of the purple cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c812d12fe8c925f72f94b05e.png"}, {"type": "image", "uri": "media/bdv2fail/f3098f1df8d5000023005fc0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0911/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_79_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 911, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_79_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "take cucumber out of cup", "subtask": "take cucumber out of cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8a661b1d19e480a6effeae34.png"}, {"type": "image", "uri": "media/bdv2fail/8f9c6ad434ecf68d7aee75c5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0912/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_take_cucumber_out_of_cup_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 912, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_take_cucumber_out_of_cup_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "move the yellow bell pepper to the left bottom corner of the table", "subtask": "move the yellow bell pepper to the left bottom corner of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/68ab763369eb2d3166866b18.png"}, {"type": "image", "uri": "media/bdv2fail/dcc1ae8f7eec4b25b5fac4bf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0913/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/43", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 913, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "43", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "take the yellow banana out of the pot and put it underneath of the table", "subtask": "take the yellow banana out of the pot and put it underneath of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fdc49c975ddf45ddac198224.png"}, {"type": "image", "uri": "media/bdv2fail/7a22268f81ad109f10c1cdc9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0914/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 914, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ef2581e8a438faa509fdffc.png"}, {"type": "image", "uri": "media/bdv2fail/e5db339e69a1888477da9d9f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0915/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 915, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "put cup to sink from drying rack", "subtask": "put cup to sink from drying rack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f75d4edf4c3579887aac356e.png"}, {"type": "image", "uri": "media/bdv2fail/4c6e6433a7469a0031e2b769.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0916/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 916, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "take the yellow brush to the lower left side of the table from the upper center.", "subtask": "take the yellow brush to the lower left side of the table from the upper center.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e347eb1a228e05441507ea5c.png"}, {"type": "image", "uri": "media/bdv2fail/5637b2725d4a5353bc7d991c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0917/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 917, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "15", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "fold the cloth from top left to bottom right", "subtask": "fold the cloth from top left to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50392ae8f1056625bc7c6066.png"}, {"type": "image", "uri": "media/bdv2fail/50392ae8f1056625bc7c6066.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0918/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_23_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 918, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_23_train", "episode_id": "8", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the green cloth to the bottom right corner of the stove", "subtask": "move the green cloth to the bottom right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a38afdf6ee867f5d8a77f241.png"}, {"type": "image", "uri": "media/bdv2fail/dc3a4c42331e380a39ced400.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0919/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_110_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 919, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_110_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the blue cup into the basket", "subtask": "put the blue cup into the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6863c54f5074103a8182f723.png"}, {"type": "image", "uri": "media/bdv2fail/618f735c173259d70592d0ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0920/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/43", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 920, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "43", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the yellow cloth to the upper right of the stove", "subtask": "move the yellow cloth to the upper right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/87883a8c33ae5d84b5115a16.png"}, {"type": "image", "uri": "media/bdv2fail/1cfcc3d8a351f7dffa3a4055.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0921/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 921, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "moove the silver pot to the upper corner of the right burner", "subtask": "moove the silver pot to the upper corner of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/084d807eb6903a08ebbce371.png"}, {"type": "image", "uri": "media/bdv2fail/36830da8da92bbc535d010a6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0922/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_61_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 922, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_61_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "place the blue spoon on the right side of the stove", "subtask": "place the blue spoon on the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/25f7eb5ce2919263dd6d6726.png"}, {"type": "image", "uri": "media/bdv2fail/a354e99bda4ac74498ef0d6a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0923/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 923, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove the green pepper from the towel", "subtask": "remove the green pepper from the towel", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f04cd54c46e4847cc8be5341.png"}, {"type": "image", "uri": "media/bdv2fail/791cd75e4d4e4b1edf8fcfe2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0924/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_05_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 924, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_05_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue fork to the bottom right corner of the stove", "subtask": "move the blue fork to the bottom right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c959092d6217d1397748c032.png"}, {"type": "image", "uri": "media/bdv2fail/96784e8df9e2a8d2c2c4b2de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0925/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 925, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the green spatula between the left and right burner", "subtask": "put the green spatula between the left and right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/89ecb4f580f85314b34d82e5.png"}, {"type": "image", "uri": "media/bdv2fail/d7ec3a2dfaf07b3dc4f6b8a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0926/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_33_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 926, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_33_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "unfold the cloth to bottom right", "subtask": "unfold the cloth to bottom right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/76972e461691977fd78ebaed.png"}, {"type": "image", "uri": "media/bdv2fail/a59c7a3447dc080e6db83246.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0927/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 927, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the stove to the bottom right of the silver pot", "subtask": "move the stove to the bottom right of the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/686d4633d579b17543c66d56.png"}, {"type": "image", "uri": "media/bdv2fail/d93db97b19b0a4b0bb68ce90.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0928/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 928, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow spoon from the left of the table to the right", "subtask": "move the yellow spoon from the left of the table to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3de0f83464560698d22cf8a4.png"}, {"type": "image", "uri": "media/bdv2fail/76567c96ef3ae8612f0c2ee5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0929/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 929, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "take the silver pot and put it to the bottom right of the burner", "subtask": "take the silver pot and put it to the bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2f0dfa89aa0ddca0ed3a5c28.png"}, {"type": "image", "uri": "media/bdv2fail/572457703fda74d124297f3c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0930/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_83_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 930, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_83_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the cucumber from the right side of the table to the left", "subtask": "move the cucumber from the right side of the table to the left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c8dddd5afa781e2031e21f66.png"}, {"type": "image", "uri": "media/bdv2fail/360b11cc7d6327ef5326f027.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0931/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 931, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_23_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "remove the blue figure from the top edge of the cloth", "subtask": "remove the blue figure from the top edge of the cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dc4228d0728db133a5314b33.png"}, {"type": "image", "uri": "media/bdv2fail/dc4228d0728db133a5314b33.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0932/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 932, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "31", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "unfold the cloth from top right to bottom left", "subtask": "unfold the cloth from top right to bottom left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec1e0594c491cad95c950b1d.png"}, {"type": "image", "uri": "media/bdv2fail/a1abbbe064b3b9d62f66c429.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0933/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_08_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 933, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_08_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove corn from bowl sink", "subtask": "remove corn from bowl sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3124b5ddd6d73d5c63413db7.png"}, {"type": "image", "uri": "media/bdv2fail/3124b5ddd6d73d5c63413db7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0934/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/1", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 934, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "1", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "fold the cloth from the left to the right", "subtask": "fold the cloth from the left to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3d11d884182e14848dbfab57.png"}, {"type": "image", "uri": "media/bdv2fail/b08c2a52061b00bab2800788.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0936/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_20_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 936, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_20_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the yellow plastic spoon to the left of the metal clamp", "subtask": "move the yellow plastic spoon to the left of the metal clamp", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/76567c96ef3ae8612f0c2ee5.png"}, {"type": "image", "uri": "media/bdv2fail/3de0f83464560698d22cf8a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0937/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 937, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "4", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "fold cloth from left to right", "subtask": "fold cloth from left to right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/34e6646aa46827c6c5853ee9.png"}, {"type": "image", "uri": "media/bdv2fail/d9b74844fa8b5123d0295782.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0938/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_20_train/31", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 938, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_20_train", "episode_id": "31", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "take the green towel and move it to the top of the desk next to the cans", "subtask": "take the green towel and move it to the top of the desk next to the cans", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9cba7b20bc5a2d9962a08ae1.png"}, {"type": "image", "uri": "media/bdv2fail/c88451ad8f5471f714fe164b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0939/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 939, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_34_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "move the red bowl to the top left corner of the stove.", "subtask": "move the red bowl to the top left corner of the stove.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b929de4c754f6db96d457213.png"}, {"type": "image", "uri": "media/bdv2fail/26e1990864c0db4bab305fc2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0940/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_40_train/8", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 940, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_40_train", "episode_id": "8", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the toy blue fork to the left side of the stove", "subtask": "move the toy blue fork to the left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1987a877a7ba672d99f9285.png"}, {"type": "image", "uri": "media/bdv2fail/ec94647865e8d103ccc9b4b0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0941/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_32_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 941, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_32_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "pick up the silver pot and placed it on the lower left side of the stove", "subtask": "pick up the silver pot and placed it on the lower left side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a32e8bf033e7e59dc22efb68.png"}, {"type": "image", "uri": "media/bdv2fail/11fdaceb7b9e5c0ca3e01c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0942/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_39_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 942, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_39_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 245, "role": "outcome"}}
{"state": {"task": "picking the silver pot from the stove and place it to the left bunner on the left side", "subtask": "picking the silver pot from the stove and place it to the left bunner on the left side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a3c3168a4bc7d69cb2264d99.png"}, {"type": "image", "uri": "media/bdv2fail/f287190cff6cff02b0cbedde.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0943/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/16", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 943, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 253, "role": "outcome"}}
{"state": {"task": "relocate the green object to the upper right side of the drawer", "subtask": "relocate the green object to the upper right side of the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5bb33868f20c6b1f0335082c.png"}, {"type": "image", "uri": "media/bdv2fail/33c436c7a52397cf29c3bb59.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0944/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/23", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 944, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "23", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "move the blue plastic fork to the toy sink", "subtask": "move the blue plastic fork to the toy sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aa67636d2834cd9744c6e590.png"}, {"type": "image", "uri": "media/bdv2fail/a9da7a85afe8afa40602f3b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0945/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 945, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the pot to the left burner.", "subtask": "move the pot to the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8c322d7c759d78937a33a526.png"}, {"type": "image", "uri": "media/bdv2fail/a2d6f6691330bcd5352cd0b9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0946/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 946, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the silver frying pan to the left of the green towel.", "subtask": "move the silver frying pan to the left of the green towel.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d48b0b79c11fbd1ce0ebbfb9.png"}, {"type": "image", "uri": "media/bdv2fail/a33bdde5f4ed17c5d274427c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0947/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 947, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_04_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put the yellow duck in the drawer", "subtask": "put the yellow duck in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/41fcb48a11b60f5deb2b499d.png"}, {"type": "image", "uri": "media/bdv2fail/9e8f843e899beaa3131cc0e7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0948/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/43", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 948, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "43", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the blue cloth to the left burner", "subtask": "move the blue cloth to the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c0aa311fd6c74ab43ee1db95.png"}, {"type": "image", "uri": "media/bdv2fail/a02ca6309886475c4cc2ee65.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0949/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 949, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_17_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move blue fork to the top right of burners", "subtask": "move blue fork to the top right of burners", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91d8e8347b3e3867bbd7d407.png"}, {"type": "image", "uri": "media/bdv2fail/91d8e8347b3e3867bbd7d407.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0950/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 950, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom right of the stove", "subtask": "move the silver pot to the bottom right of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fc45bfd4a9202a673cd53044.png"}, {"type": "image", "uri": "media/bdv2fail/7f3b33d2ac8a28fded0e56f0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0951/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_09_train/20", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 951, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_09_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cloth to the left of the table", "subtask": "move the cloth to the left of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1390895326c2191e5ae67068.png"}, {"type": "image", "uri": "media/bdv2fail/c7ffa4b258ebe4191ba03496.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0952/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train/19", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 952, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_31_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the silver pot to the lower right of the table", "subtask": "move the silver pot to the lower right of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6fc4735c434684341b575ce.png"}, {"type": "image", "uri": "media/bdv2fail/74f89f5862565443ca634df1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0953/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 953, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the yellow knife on the grey knob", "subtask": "put the yellow knife on the grey knob", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ebd528d3388bf89c6be5de3e.png"}, {"type": "image", "uri": "media/bdv2fail/97e2d93377a76d85755cd7bb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0954/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_08_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 954, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_08_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the blue laddle to the left side of the burner", "subtask": "move the blue laddle to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1537f3e7d28e2e69dbd822ba.png"}, {"type": "image", "uri": "media/bdv2fail/40b11d8a2ea75d2dd6d64aac.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0955/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_27_train/18", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 955, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_27_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the red plastic fork to the lower left side burner", "subtask": "move the red plastic fork to the lower left side burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4fbecb6bc530c1f36c14b5dc.png"}, {"type": "image", "uri": "media/bdv2fail/3779fdbb0ddbb8ec5e9cb1d1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0956/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_43_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 956, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_43_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the red fork to the right side of the stove", "subtask": "move the red fork to the right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}, {"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0957/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 957, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "10", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put the toy next to the red handle", "subtask": "put the toy next to the red handle", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a36e3a8487260877ca19c80.png"}, {"type": "image", "uri": "media/bdv2fail/8986315a67669650333507af.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0958/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train/36", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 958, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train", "episode_id": "36", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the yellow sponge to the back left from the front right corner.", "subtask": "move the yellow sponge to the back left from the front right corner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec02bd65b6583fd4c5d4b94a.png"}, {"type": "image", "uri": "media/bdv2fail/bffc7ba1b7cc212d6ed0a846.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0959/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 959, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "push the blue cloth to the upper right of the burner", "subtask": "push the blue cloth to the upper right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3977181724dce9d999d17843.png"}, {"type": "image", "uri": "media/bdv2fail/5e63ddc3c4c04ec528c27256.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0960/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 960, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_23_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot to upper right corner of the stove", "subtask": "move the silver pot to upper right corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0d3e9454203b3c40ea1b3727.png"}, {"type": "image", "uri": "media/bdv2fail/537ff4b0bca8b24112865a29.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0962/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train/9", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 962, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_18_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "remove cup from plate", "subtask": "remove cup from plate", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/05e174d4bfb7f2c026e589f6.png"}, {"type": "image", "uri": "media/bdv2fail/9eba0e1cf2b8fca02b17a05a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0963/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/30", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 963, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "30", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "move the pot lid on the pot", "subtask": "move the pot lid on the pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/83233fc8f60c7caa1b1fa065.png"}, {"type": "image", "uri": "media/bdv2fail/07ec64235cdca9c08d37ba49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0964/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_11_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 964, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_11_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "take the blue brush and put it on the left burner", "subtask": "take the blue brush and put it on the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a9d7aa88a571d3824cc66924.png"}, {"type": "image", "uri": "media/bdv2fail/e7bcb9dd50b777b9601f76b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0965/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 965, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "unfold the cloth from bottom left to top right", "subtask": "unfold the cloth from bottom left to top right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8c82a6dae95e1146a05c683c.png"}, {"type": "image", "uri": "media/bdv2fail/8c82a6dae95e1146a05c683c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0966/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_19_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 966, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_19_train", "episode_id": "6", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the croissant and put it just below the croissant", "subtask": "move the croissant and put it just below the croissant", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/46dfac3b9883fc7eefdcd120.png"}, {"type": "image", "uri": "media/bdv2fail/9b18e3b01c23b57aa149f7a4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0967/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_103_train/4", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 967, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_103_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "move the toy pan to the upper side of the table", "subtask": "move the toy pan to the upper side of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b10ba904844b8db885cc4738.png"}, {"type": "image", "uri": "media/bdv2fail/2fd61c74804401681bb1ced5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0968/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 968, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the can to the left bottom corner of the stove", "subtask": "move the can to the left bottom corner of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2664fdea7aa2c67f3fa44c03.png"}, {"type": "image", "uri": "media/bdv2fail/c92a43703038119e465cbee3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0969/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 969, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_16_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the can around but not inside the lower side of the right burner", "subtask": "move the can around but not inside the lower side of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0c8de888bf45a9546a6c42bb.png"}, {"type": "image", "uri": "media/bdv2fail/f71f45f46f18b6ded34e72f4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0970/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 970, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 241, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom left of the burner", "subtask": "move the silver pot to the bottom left of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0bba7c89149c614833390126.png"}, {"type": "image", "uri": "media/bdv2fail/0bba7c89149c614833390126.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0971/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 971, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "14", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the stove below the right side burner", "subtask": "move the stove below the right side burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4e92b3270193e689d895bc7f.png"}, {"type": "image", "uri": "media/bdv2fail/2b7f1ea7785abb36b039438d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0972/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_76_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 972, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_76_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "remove the red pepper from the silver pot", "subtask": "remove the red pepper from the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5e8b0f7b6af607747a5d36d3.png"}, {"type": "image", "uri": "media/bdv2fail/c6a49b0ab353f90c00ee1f3c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0973/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_104_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 973, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_104_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the silver pot below the white brick wall.", "subtask": "move the silver pot below the white brick wall.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9035bf4020ec0aeb2dda15ed.png"}, {"type": "image", "uri": "media/bdv2fail/4fa1fbd11168567b59be0351.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0974/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_39_train/10", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 974, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_39_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9c3a9f575abc950f561a7f4a.png"}, {"type": "image", "uri": "media/bdv2fail/e14b81cd817f306c0ab7ad40.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0976/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_23_train/17", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 976, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_23_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "take out the lid from the orange pot on and put it on the top right side of the stove", "subtask": "take out the lid from the orange pot on and put it on the top right side of the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e71a315e871c1ca9a3540ad7.png"}, {"type": "image", "uri": "media/bdv2fail/40c68c5bf217f4215c1b43b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0977/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_28_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 977, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_28_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 253, "role": "outcome"}}
{"state": {"task": "put the yellow pot to the left side of the burner", "subtask": "put the yellow pot to the left side of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e4b49fb3fa6c7a046a57d4ab.png"}, {"type": "image", "uri": "media/bdv2fail/9972bd36ea4118205cc9720c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0979/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train/15", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 979, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the silver pot to the bottom right of the burner", "subtask": "move the silver pot to the bottom right of the burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/417be5b92ba7ce76de67bba6.png"}, {"type": "image", "uri": "media/bdv2fail/4e6faddc9d115edea4ae61fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0980/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_103_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 980, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_103_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the yellow cloth behind the pot with the spatula", "subtask": "move the yellow cloth behind the pot with the spatula", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ffae550a8414fe2438918a1e.png"}, {"type": "image", "uri": "media/bdv2fail/4278f7cbaa755318dae8195c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0981/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_03_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 981, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_03_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "fold the cloth from top to bottom", "subtask": "fold the cloth from top to bottom", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d2090c366290ee5634ab062e.png"}, {"type": "image", "uri": "media/bdv2fail/d2090c366290ee5634ab062e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0982/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train/12", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 982, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_06_train", "episode_id": "12", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the red pepper to the stove", "subtask": "move the red pepper to the stove", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91eb7e3f7d640058d4a34155.png"}, {"type": "image", "uri": "media/bdv2fail/393f97f8699db3b519d02878.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0983/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_90_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 983, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_90_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put spatula beside but not on plate sink", "subtask": "put spatula beside but not on plate sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f10a3cb9e75f3eacd9db109f.png"}, {"type": "image", "uri": "media/bdv2fail/d4d70bbb1008d410fc12937a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0984/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/40", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 984, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "40", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "spatial_relation_changed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the blue object to the upper right side of the right burner", "subtask": "move the blue object to the upper right side of the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/946b113097009abfd7b9a8b2.png"}, {"type": "image", "uri": "media/bdv2fail/0bcaf4d1b95d74c883342cb3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0985/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_70_train/2", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 985, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_70_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "put the plastic cup on the drying rack", "subtask": "put the plastic cup on the drying rack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b62905c73e12d56e18b4f0e8.png"}, {"type": "image", "uri": "media/bdv2fail/ea91fa5ba5f83870104327cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0986/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train/6", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 986, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_sink_to_drying_rack_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put spatula on yellow toy", "subtask": "put spatula on yellow toy", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50f930b58ce28cc7a4868de5.png"}, {"type": "image", "uri": "media/bdv2fail/9ee6e7f6f8355b0bcfd5698b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0987/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 987, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "fold the cloth from bottom right to top left", "subtask": "fold the cloth from bottom right to top left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1298afc119b4d6883b0d5e00.png"}, {"type": "image", "uri": "media/bdv2fail/b64f2637287dbd17485d1dcf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0988/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train/0", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 988, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put cup from counter to sink", "subtask": "put cup from counter to sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a57b7b0339419b130f4f705.png"}, {"type": "image", "uri": "media/bdv2fail/0a57b7b0339419b130f4f705.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0989/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/29", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 989, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "29", "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "identical_frames"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the plastic yellow object to the left burner.", "subtask": "move the plastic yellow object to the left burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a2ab8f7cc079a4965de650f0.png"}, {"type": "image", "uri": "media/bdv2fail/b4aa86fc696697b9d4a98f84.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0990/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 990, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "remove spatula from cutting board", "subtask": "remove spatula from cutting board", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d2fdb2e842f3aa01ba3da236.png"}, {"type": "image", "uri": "media/bdv2fail/d058a69a3534df9fee6d976a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0991/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_spatula_on_cutting_board_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 991, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_put_spatula_on_cutting_board_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 225, "role": "outcome"}}
{"state": {"task": "move the orange cloth to the left burner", "subtask": "move the orange cloth to the left burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7bbd5336fef654e1c6796542.png"}, {"type": "image", "uri": "media/bdv2fail/c25b585d0fb6a22d0f397197.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0992/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_02_train/3", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 992, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_02_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the potato on the blue cloth", "subtask": "place the potato on the blue cloth", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/be4b78d65fd47c4dc8c7d7f6.png"}, {"type": "image", "uri": "media/bdv2fail/79b7e724a2b2ddb9d8f42366.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0993/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_88_train/13", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 993, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_88_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the white object inside the green object", "subtask": "put the white object inside the green object", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed6df83cafb25dade87dc4a8.png"}, {"type": "image", "uri": "media/bdv2fail/5c836022b9d3f9b4368c292a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0994/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train/14", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 994, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "move the pot above the right burner", "subtask": "move the pot above the right burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/789dd36199509b6650a94861.png"}, {"type": "image", "uri": "media/bdv2fail/6dd027b67403e580ccc0cd19.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0995/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_64_train/11", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 995, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_64_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "move the yellow sponge to the right top burner.", "subtask": "move the yellow sponge to the right top burner.", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8fd7aa3066f81bb9a4d4371d.png"}, {"type": "image", "uri": "media/bdv2fail/2144839a7e0833f955030075.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0996/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/5", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 996, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images", "construction": "instruction_perturbed"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "move the long red object to the top of the silver saucepan", "subtask": "move the long red object to the top of the silver saucepan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5166a1ef78899cb72165390.png"}, {"type": "image", "uri": "media/bdv2fail/b7cbd81851fa9794da6e4f71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0998/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_43_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 998, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_43_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 239, "role": "outcome"}}
{"state": {"task": "open the silver pot", "subtask": "open the silver pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "bdv2fail_exec_success"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1f84c9a067a132760ca03694.png"}, {"type": "image", "uri": "media/bdv2fail/7324c01c388dc225769bec0f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/exec0999/bdv2fail_exec_success", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train/7", "family": "bdv2fail_exec_success", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 999, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "Move the banana to the upper middle of the table, then to the upper left, then to the upper right, then back to the upper middle.", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana to the upper middle of the table"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped banana to the upper left of the table"}, {"step": 6, "subtask": "release banana"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped banana to the upper right of the table"}, {"step": 9, "subtask": "release banana"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped banana to the upper middle of the table"}, {"step": 12, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2a019097dd9a28e6ed69d6f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0000/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 0, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "18", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 429, "role": "error_localization"}}
{"state": {"task": "remove the lid of the silver pot and put it on the left side of the can", "plan": [{"step": 1, "subtask": "pick up silver pot lid"}, {"step": 2, "subtask": "move grasped silver pot lid away from the pot"}, {"step": 3, "subtask": "release silver pot lid"}, {"step": 4, "subtask": "move grasped silver pot lid to the left side of the can"}, {"step": 5, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/20ce42d5c4f4e70e53a66b47.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0001/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_39_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 1, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_39_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 302, "role": "error_localization"}}
{"state": {"task": "Move the yellow can to the bottom right of the table right to the silver pot, then to the top left, then back to the original spot, then under the silver pot.", "plan": [{"step": 1, "subtask": "grasp yellow can"}, {"step": 2, "subtask": "move grasped yellow can to the bottom right of the table right to the silver pot"}, {"step": 3, "subtask": "release yellow can"}, {"step": 4, "subtask": "grasp yellow can"}, {"step": 5, "subtask": "move grasped yellow can to the top left"}, {"step": 6, "subtask": "release yellow can"}, {"step": 7, "subtask": "grasp yellow can"}, {"step": 8, "subtask": "move grasped yellow can to the original spot"}, {"step": 9, "subtask": "release yellow can"}, {"step": 10, "subtask": "grasp yellow can"}, {"step": 11, "subtask": "move grasped yellow can under the silver pot"}, {"step": 12, "subtask": "release yellow can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8a85b906c16c89323f8a093e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0002/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 2, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 443, "role": "error_localization"}}
{"state": {"task": "take banana out of colander", "plan": [{"step": 1, "subtask": "pick up colander"}, {"step": 2, "subtask": "move grasped colander to a stable surface"}, {"step": 3, "subtask": "release colander"}, {"step": 4, "subtask": "pick up banana"}, {"step": 5, "subtask": "move grasped banana away from colander"}, {"step": 6, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eb1d5fd706aaab141a9771e6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0003/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train/30", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 3, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train", "episode_id": "30", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 296, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the lower right of the burner, then to the upper left, then to the center, then back to the lower right", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower right of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper left"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the lower right of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/472bc8f5862b505c2e5ce76f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0004/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_59_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 4, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_59_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 434, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from the blue cloth, place it on the wooden table, then return it to the cloth, finally move it to the green mat.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out of blue cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to wooden table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to blue cloth"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to green mat"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8ce9b9ac4ef5ac0b94b36f71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0005/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 5, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 427, "role": "error_localization"}}
{"state": {"task": "put the red pepper in the pot, then put the spatula and yellow cloth in the sink, and turn on the left burner", "plan": [{"step": 1, "subtask": "grasp red pepper"}, {"step": 2, "subtask": "move grasped red pepper to pot"}, {"step": 3, "subtask": "release red pepper"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to sink"}, {"step": 6, "subtask": "release spatula"}, {"step": 7, "subtask": "grasp yellow plate"}, {"step": 8, "subtask": "move grasped yellow plate to sink"}, {"step": 9, "subtask": "release yellow plate"}, {"step": 10, "subtask": "grasp left burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0006/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 6, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 416, "role": "error_localization"}}
{"state": {"task": "Remove the yellow scrubber from the green cloth, then place the scrubber on the table, then replace it back on the cloth.", "plan": [{"step": 1, "subtask": "grasp yellow scrubber"}, {"step": 2, "subtask": "move grasped yellow scrubber out of green cloth"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release yellow scrubber"}, {"step": 5, "subtask": "grasp yellow scrubber"}, {"step": 6, "subtask": "move grasped yellow scrubber to green cloth"}, {"step": 7, "subtask": "release yellow scrubber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c99f5cb2750d538b55f158b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0007/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 7, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "6", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 343, "role": "error_localization"}}
{"state": {"task": "Take the green pepper and place it in the plated pot, then remove the lid from the pot and place it on the table, then replace the lid on the pot.", "plan": [{"step": 1, "subtask": "grasp green pepper"}, {"step": 2, "subtask": "move grasped green pepper to plated pot"}, {"step": 3, "subtask": "release green pepper"}, {"step": 4, "subtask": "grasp pot lid"}, {"step": 5, "subtask": "move grasped pot lid away"}, {"step": 6, "subtask": "release pot lid"}, {"step": 7, "subtask": "grasp pot lid"}, {"step": 8, "subtask": "move grasped pot lid to pot"}, {"step": 9, "subtask": "release pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/eaad6b8b6b46802af4632e32.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0008/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_09_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 8, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_09_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "Move the yellow spatula to the right side of the burner, then to the front, then to the back, then to the left side.", "plan": [{"step": 1, "subtask": "grasp yellow spatula"}, {"step": 2, "subtask": "move grasped yellow spatula to the right side of the burner"}, {"step": 3, "subtask": "release yellow spatula"}, {"step": 4, "subtask": "grasp yellow spatula"}, {"step": 5, "subtask": "move grasped yellow spatula to the back"}, {"step": 6, "subtask": "release yellow spatula"}, {"step": 7, "subtask": "grasp yellow spatula"}, {"step": 8, "subtask": "move grasped yellow spatula to the left side"}, {"step": 9, "subtask": "release yellow spatula"}, {"step": 10, "subtask": "grasp yellow spatula"}, {"step": 11, "subtask": "move grasped yellow spatula to the right side of the burner"}, {"step": 12, "subtask": "release yellow spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c42b354c174671d99573e52.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0009/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 9, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "Remove the spoon from the bowl, then place it on the sink, then return it to the bowl, then place it on the counter.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to sink"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to bowl"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon to counter"}, {"step": 12, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8a58acdc503f825aff64a699.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0010/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/29", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 10, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "29", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "remove the green object from between the burners, then place it on the table, then return it to the original spot, then place it beside the stove", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object out from between the burners"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object to table"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "grasp green object"}, {"step": 8, "subtask": "move grasped green object to original spot"}, {"step": 9, "subtask": "release green object"}, {"step": 10, "subtask": "grasp green object"}, {"step": 11, "subtask": "move grasped green object beside the stove"}, {"step": 12, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6ea37b21854ef0c98071472d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0011/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 11, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 429, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the stove, then to the back of the stove, then to the front of the stove.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the front of the stove"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7ab2ec5c60b65199ea5eabb9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0012/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_25_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 12, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_25_train", "episode_id": "12", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 317, "role": "error_localization"}}
{"state": {"task": "put cup from counter to sink, then from sink to dishwasher", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to sink"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to dishwasher"}, {"step": 6, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc5ec89012aeb7227cfd52ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0013/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 13, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 295, "role": "error_localization"}}
{"state": {"task": "move the green object in the pot above the right burner.", "plan": [{"step": 1, "subtask": "pick up green object"}, {"step": 2, "subtask": "release green object"}, {"step": 3, "subtask": "pick up pot"}, {"step": 4, "subtask": "move grasped pot above the right burner"}, {"step": 5, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd2bd32f33d4e6aa80411a08.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0014/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_86_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 14, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_86_train", "episode_id": "11", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 280, "role": "error_localization"}}
{"state": {"task": "move the knob-like object to the left side of the table, then to the right side, then back to the left side", "plan": [{"step": 1, "subtask": "grasp knob-like object"}, {"step": 2, "subtask": "move grasped knob-like object to the left side of the table"}, {"step": 3, "subtask": "release knob-like object"}, {"step": 4, "subtask": "grasp knob-like object"}, {"step": 5, "subtask": "move grasped knob-like object to the right side of the table"}, {"step": 6, "subtask": "release knob-like object"}, {"step": 7, "subtask": "grasp knob-like object"}, {"step": 8, "subtask": "move grasped knob-like object to the left side of the table"}, {"step": 9, "subtask": "release knob-like object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ecdfd201f556737580f33240.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0015/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 15, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 388, "role": "error_localization"}}
{"state": {"task": "Remove the orange cloth from the left stove, then fold it neatly, place it on the counter, then return it to the stove.", "plan": [{"step": 1, "subtask": "grasp orange cloth"}, {"step": 2, "subtask": "move grasped orange cloth away from left stove"}, {"step": 3, "subtask": "release orange cloth"}, {"step": 4, "subtask": "grasp orange cloth"}, {"step": 5, "subtask": "rotate grasped orange cloth"}, {"step": 6, "subtask": "move grasped orange cloth to counter"}, {"step": 7, "subtask": "release orange cloth"}, {"step": 8, "subtask": "grasp orange cloth"}, {"step": 9, "subtask": "move grasped orange cloth to left stove"}, {"step": 10, "subtask": "release orange cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6eba7c890cadea21880a9e77.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0016/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 16, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 386, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the bottom right corner of the stove, then to the top left corner, then to the bottom left corner, then back to the bottom right corner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom right of the stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top left of the stove"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the bottom left of the stove"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the bottom right of the stove"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d4a9b4c5e1e4c843da67f4f0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0017/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 17, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 446, "role": "error_localization"}}
{"state": {"task": "Put the cup on the plate, then lift the cup off the plate, then put it back on the plate.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to plate"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup up"}, {"step": 6, "subtask": "grasp cup"}, {"step": 7, "subtask": "move grasped cup to plate"}, {"step": 8, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9eba0e1cf2b8fca02b17a05a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0018/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/30", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 18, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "30", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 337, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl, then take bowl out of sink onto the table, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "move grasped bowl to the table"}, {"step": 7, "subtask": "release bowl"}, {"step": 8, "subtask": "grasp cabinet door handle"}, {"step": 9, "subtask": "move grasped door handle to open position"}, {"step": 10, "subtask": "release door handle"}, {"step": 11, "subtask": "grasp strawberry"}, {"step": 12, "subtask": "move grasped strawberry to oven"}, {"step": 13, "subtask": "release strawberry"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0019/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 19, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 433, "role": "error_localization"}}
{"state": {"task": "Put cup from drying rack to sink, then place it on countertop, then return it to drying rack.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to sink"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to countertop"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to drying rack"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0da76b0adfe9e1e765998e61.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0020/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 20, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "place the brush on the purple cloth, then on the green mat, then back on the purple cloth", "plan": [{"step": 1, "subtask": "grasp brush"}, {"step": 2, "subtask": "move grasped brush to purple cloth"}, {"step": 3, "subtask": "release brush"}, {"step": 4, "subtask": "grasp brush"}, {"step": 5, "subtask": "move grasped brush to green mat"}, {"step": 6, "subtask": "release brush"}, {"step": 7, "subtask": "grasp brush"}, {"step": 8, "subtask": "move grasped brush to purple cloth"}, {"step": 9, "subtask": "release brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8890b949807b6acc0245e76b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0021/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_30_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 21, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_30_train", "episode_id": "3", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "Move the red fork to the right side of the stove, then to the left side, then back to the right side.", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "release red fork"}, {"step": 5, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 6, "subtask": "release red fork"}, {"step": 7, "subtask": "grasp red fork"}, {"step": 8, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 9, "subtask": "grasp red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0022/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 22, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "10", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 378, "role": "error_localization"}}
{"state": {"task": "take the fish out of the pan and put it on the lower right side of the stove", "plan": [{"step": 1, "subtask": "pick up fish"}, {"step": 2, "subtask": "move grasped fish away from pan"}, {"step": 3, "subtask": "release fish"}, {"step": 4, "subtask": "move grasped fish to the lower right side of the stove"}, {"step": 5, "subtask": "release fish"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/381c5840e7c0aa66247b2755.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0023/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_27_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 23, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_27_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 293, "role": "error_localization"}}
{"state": {"task": "take the yellow banana and put in the left bottom of the table, then the apple next to it, then the orange above the banana", "plan": [{"step": 1, "subtask": "grasp yellow banana"}, {"step": 2, "subtask": "move grasped yellow banana to the left bottom of the table"}, {"step": 3, "subtask": "release yellow banana"}, {"step": 4, "subtask": "grasp apple"}, {"step": 5, "subtask": "move grasped apple next to the yellow banana"}, {"step": 6, "subtask": "release apple"}, {"step": 7, "subtask": "grasp orange"}, {"step": 8, "subtask": "move grasped orange above the yellow banana"}, {"step": 9, "subtask": "release orange"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a61f3f332984fa96881c32be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0024/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train/26", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 24, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_22_train", "episode_id": "26", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 370, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tap handle"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp bowl"}, {"step": 10, "subtask": "move grasped bowl to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp left burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0025/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 25, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 579, "role": "error_localization"}}
{"state": {"task": "Move one of the cans to the upper left corner of the left burner, then to the upper right corner, then to the lower left corner, and finally to the lower right corner.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the upper left corner of the left burner"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the upper right corner of the left burner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the lower left corner of the left burner"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "move grasped can to the lower right corner of the left burner"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c850847225497b2c7a0b7727.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0026/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 26, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 444, "role": "error_localization"}}
{"state": {"task": "put the item closest to the center of the right burner into the pot", "plan": [{"step": 1, "subtask": "grasp green towel"}, {"step": 2, "subtask": "move grasped green towel into the pot"}, {"step": 3, "subtask": "release green towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0027/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 27, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 254, "role": "error_localization"}}
{"state": {"task": "remove the yellow object from th ecloth in", "plan": [{"step": 1, "subtask": "pick up yellow object"}, {"step": 2, "subtask": "move grasped yellow object away from cloth"}, {"step": 3, "subtask": "put it back where it was"}, {"step": 4, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/18ce7a1e0faa7922f55bf051.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0028/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 28, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 267, "role": "error_localization"}}
{"state": {"task": "put red pot on the right burner, add the tomato and cheese into it, turn on the corresponding burner", "plan": [{"step": 1, "subtask": "grasp red pot"}, {"step": 2, "subtask": "move grasped red pot to right burner"}, {"step": 3, "subtask": "release red pot"}, {"step": 4, "subtask": "grasp tomato"}, {"step": 5, "subtask": "move grasped tomato into red pot"}, {"step": 6, "subtask": "release tomato"}, {"step": 7, "subtask": "grasp cheese"}, {"step": 8, "subtask": "move grasped cheese into red pot"}, {"step": 9, "subtask": "release cheese"}, {"step": 10, "subtask": "grasp left burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0029/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 29, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 408, "role": "error_localization"}}
{"state": {"task": "place the green object inside the drawer", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up green object"}, {"step": 3, "subtask": "move grasped green object to drawer"}, {"step": 4, "subtask": "release green object"}, {"step": 5, "subtask": "close drawer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e91d0dd17e8e5cae5243902a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0030/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 30, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_01_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 274, "role": "error_localization"}}
{"state": {"task": "place the spoon at the left side of the stove", "plan": [{"step": 1, "subtask": "pick up spoon"}, {"step": 2, "subtask": "move grasped spoon to the left side of the stove"}, {"step": 3, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8d3824123baf606e3bbcc846.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0031/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_35_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 31, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_35_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 251, "role": "error_localization"}}
{"state": {"task": "heat the bread for 25 minutes in the oven, then put it back in the orange bowl", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl to oven"}, {"step": 6, "subtask": "release bowl"}, {"step": 7, "subtask": "grasp oven door handle"}, {"step": 8, "subtask": "move grasped door handle to close position"}, {"step": 9, "subtask": "release door handle"}, {"step": 10, "subtask": "wait 25 minutes"}, {"step": 11, "subtask": "grasp door handle"}, {"step": 12, "subtask": "move grasped door handle to open position"}, {"step": 13, "subtask": "release door handle"}, {"step": 14, "subtask": "grasp bread"}, {"step": 15, "subtask": "move grasped bread to orange bowl"}, {"step": 16, "subtask": "release bread"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0032/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 32, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 483, "role": "error_localization"}}
{"state": {"task": "move the slice of orange on the side of the table, then move it to the center, then move it back to the side, then pick up another slice and place it next to the first", "plan": [{"step": 1, "subtask": "grasp orange slice"}, {"step": 2, "subtask": "move grasped orange slice to the side of the table"}, {"step": 3, "subtask": "release orange slice"}, {"step": 4, "subtask": "grasp orange slice"}, {"step": 5, "subtask": "move grasped orange slice to the center of the table"}, {"step": 6, "subtask": "release orange slice"}, {"step": 7, "subtask": "grasp another orange slice"}, {"step": 8, "subtask": "move grasped orange slice next to the first orange slice"}, {"step": 9, "subtask": "release orange slice"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/082fb7b21b913ced4428ff74.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0033/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 33, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "19", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 391, "role": "error_localization"}}
{"state": {"task": "move the silver pot between two burners, then return it to the starting burner, then move it back again", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the second burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the starting burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the second burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0034/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 34, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 367, "role": "error_localization"}}
{"state": {"task": "remove the lid of the silver pot and put it on the left side of the can", "plan": [{"step": 1, "subtask": "pick up silver pot lid"}, {"step": 2, "subtask": "move grasped silver pot lid away from the pot"}, {"step": 3, "subtask": "release silver pot lid"}, {"step": 4, "subtask": "move grasped silver pot lid to the right side of the can"}, {"step": 5, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/20ce42d5c4f4e70e53a66b47.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0035/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_39_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 35, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_39_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 302, "role": "error_localization"}}
{"state": {"task": "Take the potato, move it to the right side of the stove, then to the left, then back to the right.", "plan": [{"step": 1, "subtask": "grasp potato"}, {"step": 2, "subtask": "move grasped potato to the right side of the stove"}, {"step": 3, "subtask": "release potato"}, {"step": 4, "subtask": "move grasped potato to the left side of the stove"}, {"step": 5, "subtask": "release potato"}, {"step": 6, "subtask": "grasp potato"}, {"step": 7, "subtask": "move grasped potato to the right side of the stove"}, {"step": 8, "subtask": "release potato"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0036/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 36, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 354, "role": "error_localization"}}
{"state": {"task": "Pick up the blue spoon, place it on the orange towel, then move the spoon to the green plate, and finally place it back on the orange towel.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to orange towel"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to green plate"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to orange towel"}, {"step": 9, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8b9d1710a803c686409d55f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0037/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_06_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 37, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_06_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 373, "role": "error_localization"}}
{"state": {"task": "Remove the spoon from the orange pot, then place it on the table, pick it up again, and return it to the orange pot.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of orange pot"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to table"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to orange pot"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4b15d6af50f2480076658224.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0038/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 38, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "move silver pot over orange cloth on the right stove, then the left stove, then back to the right stove", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot over orange cloth on the right stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "move grasped silver pot to the left stove"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot over orange cloth on the right stove"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/28ccd04f674bb1bce1a2ad4b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0039/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 39, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 389, "role": "error_localization"}}
{"state": {"task": "move the yellow knife to the bottom middle part of the stove, then to the top left, then to the top right, then back to the bottom middle", "plan": [{"step": 1, "subtask": "grasp yellow knife"}, {"step": 2, "subtask": "move grasped yellow knife to the bottom middle part of the stove"}, {"step": 3, "subtask": "release yellow knife"}, {"step": 4, "subtask": "grasp yellow knife"}, {"step": 5, "subtask": "move grasped yellow knife to the top left of the stove"}, {"step": 6, "subtask": "release yellow knife"}, {"step": 7, "subtask": "grasp yellow knife"}, {"step": 8, "subtask": "move grasped yellow knife to the top right of the stove"}, {"step": 9, "subtask": "release yellow knife"}, {"step": 10, "subtask": "grasp yellow knife"}, {"step": 11, "subtask": "move grasped yellow knife to the bottom middle part of the stove"}, {"step": 12, "subtask": "release yellow knife"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ff76cb4da0d554edd36bd7c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0040/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train/20", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 40, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_29_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "place the eggplant on top of the green cloth, then move the cloth with the eggplant to the table, then remove the eggplant and place it on the floor", "plan": [{"step": 1, "subtask": "grasp eggplant"}, {"step": 2, "subtask": "move grasped eggplant to top of the green cloth"}, {"step": 3, "subtask": "release eggplant"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth to the table"}, {"step": 6, "subtask": "release green cloth"}, {"step": 7, "subtask": "grasp eggplant"}, {"step": 8, "subtask": "move grasped eggplant to the floor"}, {"step": 9, "subtask": "release eggplant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2598fd4c6c2589d34bf8a1d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0041/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_39_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 41, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_39_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 379, "role": "error_localization"}}
{"state": {"task": "move the white and orange object on the right burner", "plan": [{"step": 1, "subtask": "pick up white object"}, {"step": 2, "subtask": "move grasped white object to right burner"}, {"step": 3, "subtask": "release white object"}, {"step": 4, "subtask": "pick up orange object"}, {"step": 5, "subtask": "move grasped orange object to right burner"}, {"step": 6, "subtask": "release orange object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d2041f4c15a4c9033cbdba7d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0042/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 42, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 301, "role": "error_localization"}}
{"state": {"task": "open the oven, then move the pot to the open oven door, then close the oven, then move the pan to the open oven door", "plan": [{"step": 1, "subtask": "grasp oven handle"}, {"step": 2, "subtask": "move grasped oven handle to oven's open position"}, {"step": 3, "subtask": "release oven handle"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to open oven door"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp oven handle"}, {"step": 8, "subtask": "move grasped oven handle to oven's closed position"}, {"step": 9, "subtask": "release oven handle"}, {"step": 10, "subtask": "grasp pan"}, {"step": 11, "subtask": "move grasped pan to open oven door"}, {"step": 12, "subtask": "release pan"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3b82692e6e4e5950f12ad350.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0043/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_close_oven_train/38", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 43, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_close_oven_train", "episode_id": "38", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 424, "role": "error_localization"}}
{"state": {"task": "move the can to the top of the left burner, then to the right burner, then back to the left burner.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the top of the left burner"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the right burner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the top of the left burner"}, {"step": 9, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d8fe49c3e772f22e6f6be79c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0044/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 44, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 365, "role": "error_localization"}}
{"state": {"task": "place the blue brush on top of the purple cloth", "plan": [{"step": 1, "subtask": "pick up blue brush"}, {"step": 2, "subtask": "move grasped blue brush to top of the purple cloth"}, {"step": 3, "subtask": "release blue brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c812d12fe8c925f72f94b05e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0045/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_79_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 45, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_79_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 253, "role": "error_localization"}}
{"state": {"task": "Move the yellow object to the bottom right edge of burners, then to the top left edge, then back to the bottom right edge.", "plan": [{"step": 1, "subtask": "grasp yellow object"}, {"step": 2, "subtask": "move grasped yellow object to bottom right edge of burners"}, {"step": 3, "subtask": "release yellow object"}, {"step": 4, "subtask": "grasp yellow object"}, {"step": 5, "subtask": "move grasped yellow object to top left edge of burners"}, {"step": 6, "subtask": "release yellow object"}, {"step": 7, "subtask": "grasp yellow object"}, {"step": 8, "subtask": "move grasped yellow object to bottom right edge of burners"}, {"step": 9, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5256d032f07b2e7b5bb8362c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0046/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 46, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 381, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth to the bottom right corner of the table, then fold it, then unfold it, then move it to the top left corner.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to the bottom right corner of the table"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth forward"}, {"step": 6, "subtask": "move grasped yellow cloth to folded position"}, {"step": 7, "subtask": "release yellow cloth"}, {"step": 8, "subtask": "grasp yellow cloth"}, {"step": 9, "subtask": "move grasped yellow cloth from folded position"}, {"step": 10, "subtask": "move grasped yellow cloth to folded position"}, {"step": 11, "subtask": "release yellow cloth"}, {"step": 12, "subtask": "grasp yellow cloth"}, {"step": 13, "subtask": "move grasped yellow cloth to the bottom right corner of the table"}, {"step": 14, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ab4ea075c00e99b119946741.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0047/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_04_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 47, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_04_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 475, "role": "error_localization"}}
{"state": {"task": "move the pot above the left burner.", "plan": [{"step": 1, "subtask": "pick up pot"}, {"step": 2, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dbc84c7a8bc41a802ad66bab.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0048/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_76_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 48, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_76_train", "episode_id": "15", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the lower left side of the burner, then to the lower right side, then back to the lower left side", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower left side of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "move grasped silver pot to the lower right side of the burner"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to the lower left side of the burner"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/88f022e072caaccb232ef323.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0049/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_28_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 49, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_28_train", "episode_id": "0", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 400, "role": "error_localization"}}
{"state": {"task": "Put spoon in bowl, then in sink, then back in bowl, then move bowl to the counter.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "release spoon"}, {"step": 3, "subtask": "move grasped spoon to bowl"}, {"step": 4, "subtask": "release spoon"}, {"step": 5, "subtask": "grasp spoon"}, {"step": 6, "subtask": "move grasped spoon to sink"}, {"step": 7, "subtask": "release spoon"}, {"step": 8, "subtask": "grasp spoon"}, {"step": 9, "subtask": "move grasped spoon to bowl"}, {"step": 10, "subtask": "release spoon"}, {"step": 11, "subtask": "grasp bowl"}, {"step": 12, "subtask": "move grasped bowl to counter"}, {"step": 13, "subtask": "release bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ad5a66ec02e008736204152d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0050/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/25", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 50, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "25", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 415, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the lower right of the burner, then to the upper left, then to the center, then back to the lower right", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower right of the burner"}, {"step": 3, "subtask": "move grasped silver pot to the center"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper left"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "release silver pot"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the lower right of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/472bc8f5862b505c2e5ce76f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0051/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_59_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 51, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_59_train", "episode_id": "5", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 434, "role": "error_localization"}}
{"state": {"task": "take spatula off plate sink, then place it on the counter, then pick it up again, then return it to the plate sink", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula out"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to counter"}, {"step": 6, "subtask": "release spatula"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to plate sink"}, {"step": 9, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/09c16a5ebc339db4921d2890.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0052/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spatula_off_plate_sink_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 52, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spatula_off_plate_sink_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the left, then to the right, then to the center, then back to the left.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the right"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the left"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2eafcc3f152844874a2521a5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0053/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 53, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 419, "role": "error_localization"}}
{"state": {"task": "move the green object in front of the pot, then behind the pot, then back in front", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object in front of the pot"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object behind the pot"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/81bf1afcd54f4294f5d8c347.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0054/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_34_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 54, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_34_train", "episode_id": "15", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 327, "role": "error_localization"}}
{"state": {"task": "Move the yellow can to the bottom right of the table right to the silver pot, then to the top left, then back to the original spot, then under the silver pot.", "plan": [{"step": 1, "subtask": "grasp yellow can"}, {"step": 2, "subtask": "move grasped yellow can to the bottom right of the table right to the silver pot"}, {"step": 3, "subtask": "release yellow can"}, {"step": 4, "subtask": "grasp yellow can"}, {"step": 5, "subtask": "move grasped yellow can to the top left"}, {"step": 6, "subtask": "release yellow can"}, {"step": 7, "subtask": "grasp yellow can"}, {"step": 8, "subtask": "move grasped yellow can to the original spot"}, {"step": 9, "subtask": "release yellow can"}, {"step": 10, "subtask": "grasp yellow can"}, {"step": 11, "subtask": "move grasped yellow can under the silver pot"}, {"step": 12, "subtask": "put it back where it was"}, {"step": 13, "subtask": "release yellow can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8a85b906c16c89323f8a093e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0055/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 55, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 462, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the burner from the top of a blue cloth, then move it to the sink, then return it to the cloth.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out"}, {"step": 3, "subtask": "move grasped silver pot to burner"}, {"step": 4, "subtask": "release silver pot"}, {"step": 5, "subtask": "grasp silver pot"}, {"step": 6, "subtask": "move grasped silver pot to sink"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to blue cloth"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bffebe9b7211536c2dc094b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0056/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 56, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "take the silver lid off the silver pot and put it on the left side of the silver pot", "plan": [{"step": 1, "subtask": "pick up silver lid"}, {"step": 2, "subtask": "release silver lid"}, {"step": 3, "subtask": "move grasped silver lid away from silver pot"}, {"step": 4, "subtask": "release silver lid"}, {"step": 5, "subtask": "move grasped silver lid to the left side of the silver pot"}, {"step": 6, "subtask": "release silver lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e07266de493f89fdd46818e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0057/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 57, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 315, "role": "error_localization"}}
{"state": {"task": "pick up scissors and put into drawer", "plan": [{"step": 1, "subtask": "pick up drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "release scissors"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e965f9d8e06b966ff135b275.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0058/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train/38", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 58, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train", "episode_id": "38", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 263, "role": "error_localization"}}
{"state": {"task": "the robot arm moved the corn from the stove to the counter top, then back to the stove, then to the counter top again", "plan": [{"step": 1, "subtask": "grasp corn"}, {"step": 2, "subtask": "move grasped corn from stove to counter top"}, {"step": 3, "subtask": "release corn"}, {"step": 4, "subtask": "grasp corn"}, {"step": 5, "subtask": "move grasped corn from counter top to stove"}, {"step": 6, "subtask": "release corn"}, {"step": 7, "subtask": "grasp corn"}, {"step": 8, "subtask": "move grasped corn from stove to counter top"}, {"step": 9, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac8df1d3a048133c98a7aacd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0059/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 59, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 365, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from the orange cloth, place it on the table, then return it to the orange cloth, then move it to the green towel.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out of orange cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp orange towel"}, {"step": 8, "subtask": "move grasped silver pot to orange cloth"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the green towel"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0060/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 60, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 425, "role": "error_localization"}}
{"state": {"task": "move blue doll from top of the drawer and place it on the table", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up blue doll"}, {"step": 3, "subtask": "move grasped blue doll to table"}, {"step": 4, "subtask": "release blue doll"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ad6671a2052a36849b18cad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0061/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 61, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 267, "role": "error_localization"}}
{"state": {"task": "Put the spoon in the orange pot, then remove the lid from the pot, then put the spoon back in the pot.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to orange pot"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp pot lid"}, {"step": 5, "subtask": "move grasped pot lid away"}, {"step": 6, "subtask": "release pot lid"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to orange pot"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aed4303462eb9330157daa5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0062/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 62, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "put blueberries on plate sink", "plan": [{"step": 1, "subtask": "pick up blueberries"}, {"step": 2, "subtask": "move grasped blueberries to plate"}, {"step": 3, "subtask": "release blueberries"}, {"step": 4, "subtask": "pick up plate"}, {"step": 5, "subtask": "move grasped plate to sink"}, {"step": 6, "subtask": "release plate"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8557e6bc8019e1d704254a37.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0063/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 63, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 292, "role": "error_localization"}}
{"state": {"task": "put the item closest to the sink into the sink", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the sink"}, {"step": 3, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0064/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 64, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "Move the stainless pot to the upper right side of the burner, then to the lower left side, then back to the upper right side", "plan": [{"step": 1, "subtask": "grasp stainless pot"}, {"step": 2, "subtask": "grasp stainless pot"}, {"step": 3, "subtask": "release stainless pot"}, {"step": 4, "subtask": "move grasped stainless pot to the upper right side of the burner"}, {"step": 5, "subtask": "move grasped stainless pot to the lower left side"}, {"step": 6, "subtask": "release stainless pot"}, {"step": 7, "subtask": "grasp stainless pot"}, {"step": 8, "subtask": "move grasped stainless pot to the upper right side of the burner"}, {"step": 9, "subtask": "release stainless pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2ad53c280039412e29d56b31.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0065/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 65, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "7", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 381, "role": "error_localization"}}
{"state": {"task": "remove the violet cloth from the bottom right corner of the stove, then fold it neatly, place it on the countertop, then put it back onto the stove", "plan": [{"step": 1, "subtask": "grasp violet cloth"}, {"step": 2, "subtask": "move grasped violet cloth away from the bottom right corner of the stove"}, {"step": 3, "subtask": "release violet cloth"}, {"step": 4, "subtask": "grasp violet cloth"}, {"step": 5, "subtask": "fold grasped violet cloth"}, {"step": 6, "subtask": "move grasped violet cloth to the floor"}, {"step": 7, "subtask": "release violet cloth"}, {"step": 8, "subtask": "grasp violet cloth"}, {"step": 9, "subtask": "move grasped violet cloth to bottom right corner of the stove"}, {"step": 10, "subtask": "release violet cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1463de8f48b1d9c93229d0de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0066/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_84_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 66, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_84_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 402, "role": "error_localization"}}
{"state": {"task": "remove cucumber from cup, then place cup on table, then pick up cucumber and place it next to the cup", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber out of cup"}, {"step": 3, "subtask": "release cucumber"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to table"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cucumber"}, {"step": 8, "subtask": "move grasped cucumber next to cup"}, {"step": 9, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/195f3981ca5b5447204a81ce.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0067/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 67, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 354, "role": "error_localization"}}
{"state": {"task": "Put the gray stuffed animal on the table, then to the left, then to the right, then back to center.", "plan": [{"step": 1, "subtask": "grasp gray stuffed animal"}, {"step": 2, "subtask": "move grasped gray stuffed animal to table"}, {"step": 3, "subtask": "release gray stuffed animal"}, {"step": 4, "subtask": "grasp gray stuffed animal"}, {"step": 5, "subtask": "move grasped gray stuffed animal to the left"}, {"step": 6, "subtask": "release gray stuffed animal"}, {"step": 7, "subtask": "grasp gray stuffed animal"}, {"step": 8, "subtask": "move grasped gray stuffed animal to the right"}, {"step": 9, "subtask": "release gray stuffed animal"}, {"step": 10, "subtask": "grasp gray stuffed animal"}, {"step": 11, "subtask": "move grasped gray stuffed animal to center"}, {"step": 12, "subtask": "release gray stuffed animal"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c3fd255d2a035de8e420231.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0068/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_15_train/33", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 68, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_15_train", "episode_id": "33", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 429, "role": "error_localization"}}
{"state": {"task": "Move the purple cloth to the bottom right corner of the stove, then fold it, then unfold it, then move it to the bottom left corner of the stove.", "plan": [{"step": 1, "subtask": "grasp purple cloth"}, {"step": 2, "subtask": "move grasped purple cloth to the bottom right corner of the stove"}, {"step": 3, "subtask": "release purple cloth"}, {"step": 4, "subtask": "grasp purple cloth"}, {"step": 5, "subtask": "move grasped purple cloth to folded position"}, {"step": 6, "subtask": "release purple cloth"}, {"step": 7, "subtask": "grasp purple cloth"}, {"step": 8, "subtask": "move grasped purple cloth to unfolded position"}, {"step": 9, "subtask": "release purple cloth"}, {"step": 10, "subtask": "grasp purple cloth"}, {"step": 11, "subtask": "move grasped purple cloth to the bottom left corner of the stove"}, {"step": 12, "subtask": "release purple cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/188043555a15ef77efd456e0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0069/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_106_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 69, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_106_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 438, "role": "error_localization"}}
{"state": {"task": "put cup on plate, then place plate on table, pick cup off plate, put cup on floor", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to plate"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to floor"}, {"step": 6, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b149696486aae0c56d3bc369.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0070/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 70, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 303, "role": "error_localization"}}
{"state": {"task": "Move the silver pot on the lower left corner side of the cooker, then to the upper right corner, then back to the lower left corner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper left corner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the lower right corner"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c879f03ee5ab220de74086d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0071/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_77_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 71, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_77_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 323, "role": "error_localization"}}
{"state": {"task": "move the blue silver spoon from left side of the burner to the center of the induction stove, then to the right side, then back to center.", "plan": [{"step": 1, "subtask": "grasp blue silver spoon"}, {"step": 2, "subtask": "move grasped blue silver spoon to the center of the induction stove"}, {"step": 3, "subtask": "release blue silver spoon"}, {"step": 4, "subtask": "grasp blue silver spoon"}, {"step": 5, "subtask": "move grasped blue silver spoon to the right side of the induction stove"}, {"step": 6, "subtask": "release blue silver spoon"}, {"step": 7, "subtask": "grasp blue silver spoon"}, {"step": 8, "subtask": "move grasped blue silver spoon to the center of the induction stove"}, {"step": 9, "subtask": "release blue silver spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3cd21797217244e8c8f8aa71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0072/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_49_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 72, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_49_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 393, "role": "error_localization"}}
{"state": {"task": "put cucumber in cup, then take cucumber out of cup, then put cucumber in cup again", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "release cucumber"}, {"step": 3, "subtask": "grasp cucumber"}, {"step": 4, "subtask": "move grasped cucumber out of cup"}, {"step": 5, "subtask": "release cucumber"}, {"step": 6, "subtask": "grasp cucumber"}, {"step": 7, "subtask": "move grasped cucumber to cup"}, {"step": 8, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5fbbaf7e9267c641569720e1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0073/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 73, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 331, "role": "error_localization"}}
{"state": {"task": "Move the silver bowl to the upper left side of the stove, then to the upper right side, then to the lower left side, then to the lower right side.", "plan": [{"step": 1, "subtask": "grasp silver bowl"}, {"step": 2, "subtask": "move grasped silver bowl to the upper left side of the stove"}, {"step": 3, "subtask": "release silver bowl"}, {"step": 4, "subtask": "grasp silver bowl"}, {"step": 5, "subtask": "move grasped silver bowl to the upper right side of the stove"}, {"step": 6, "subtask": "release silver bowl"}, {"step": 7, "subtask": "grasp silver bowl"}, {"step": 8, "subtask": "move grasped silver bowl to the lower left side of the stove"}, {"step": 9, "subtask": "release silver bowl"}, {"step": 10, "subtask": "grasp silver bowl"}, {"step": 11, "subtask": "move grasped silver bowl to the lower right side of the stove"}, {"step": 12, "subtask": "put it back where it was"}, {"step": 13, "subtask": "release silver bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7faa098ac776b0b560cb4ef6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0074/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_19_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 74, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_19_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 468, "role": "error_localization"}}
{"state": {"task": "Open the pan lid, place the spoon inside, then remove the spoon, close the pan lid, and place the fork next to it.", "plan": [{"step": 1, "subtask": "grasp pan lid"}, {"step": 2, "subtask": "move grasped pan lid upwards"}, {"step": 3, "subtask": "release pan lid"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to inside pan"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon out of pan"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp pan lid"}, {"step": 11, "subtask": "move grasped pan lid downwards"}, {"step": 12, "subtask": "release pan lid"}, {"step": 13, "subtask": "grasp fork"}, {"step": 14, "subtask": "move grasped fork next to pan lid"}, {"step": 15, "subtask": "move it back to the original position"}, {"step": 16, "subtask": "release fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/19ddcb9d3b683562792bb23e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0075/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 75, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train", "episode_id": "5", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 485, "role": "error_localization"}}
{"state": {"task": "move the can to the bottom below the right burner, then to the top above the left burner, then back to the bottom below the right burner", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the bottom below the right burner"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the top above the left burner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the bottom below the right burner"}, {"step": 9, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6d88ea8e0618fb38ebcd88cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0076/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 76, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 374, "role": "error_localization"}}
{"state": {"task": "Take the silver vessel and place it on the purple cloth, then remove a cup from the box, place the cup on the cloth, then move the vessel to the cup.", "plan": [{"step": 1, "subtask": "grasp silver vessel"}, {"step": 2, "subtask": "move grasped silver vessel to purple cloth"}, {"step": 3, "subtask": "release silver vessel"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to purple cloth"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp silver vessel"}, {"step": 8, "subtask": "move grasped silver vessel to cup"}, {"step": 9, "subtask": "release silver vessel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/62f99b51f3c8b8557574d4a6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0077/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_tabletop_dark_wood_pnp_sweep_02_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 77, "taskvar": "numpy_256_rss_tabletop_dark_wood_pnp_sweep_02_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "take banana out of colander", "plan": [{"step": 1, "subtask": "pick up colander"}, {"step": 2, "subtask": "move grasped colander to a stable surface"}, {"step": 3, "subtask": "release colander"}, {"step": 4, "subtask": "pick up banana"}, {"step": 5, "subtask": "move grasped banana away from colander"}, {"step": 6, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/319a91a211ff5ceb9d9d0efc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0078/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train/20", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 78, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 296, "role": "error_localization"}}
{"state": {"task": "Put the blue cloth on the top edge of the cloth, then on the bottom edge, then the left edge, then the right edge.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth to the top edge of the cloth"}, {"step": 3, "subtask": "release blue cloth"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "release blue cloth"}, {"step": 6, "subtask": "grasp blue cloth"}, {"step": 7, "subtask": "move grasped blue cloth to the left edge of the cloth"}, {"step": 8, "subtask": "release blue cloth"}, {"step": 9, "subtask": "grasp blue cloth"}, {"step": 10, "subtask": "move grasped blue cloth to the right edge of the cloth"}, {"step": 11, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4ecd811048f34b0d2746cd4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0079/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 79, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 414, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the can that is the closest to the microwave", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp yellow can"}, {"step": 11, "subtask": "move grasped yellow can to upper right corner"}, {"step": 12, "subtask": "release yellow can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0080/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 80, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 436, "role": "error_localization"}}
{"state": {"task": "Put the cup from the counter to the sink, then to the dish rack, then back to the counter", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "release cup"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to dish rack"}, {"step": 6, "subtask": "move grasped cup to sink"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to counter"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/47fae74b78086735b8de3be0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0081/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/52", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 81, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "52", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "put cup to counter from sink, then take another cup from the counter to the sink, then put the first cup back to the sink", "plan": [{"step": 1, "subtask": "move grasped cup to sink"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp another cup"}, {"step": 5, "subtask": "grasp cup"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to sink"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d025406977e108f8b7331708.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0082/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/70", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 82, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "70", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "remove cup from plate, place it on the table, then move the plate to the sink, then return the cup to the plate", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "return it to its original position"}, {"step": 4, "subtask": "release cup"}, {"step": 5, "subtask": "grasp cup"}, {"step": 6, "subtask": "move grasped cup to table"}, {"step": 7, "subtask": "release cup"}, {"step": 8, "subtask": "grasp plate"}, {"step": 9, "subtask": "move grasped plate to sink"}, {"step": 10, "subtask": "release plate"}, {"step": 11, "subtask": "grasp cup"}, {"step": 12, "subtask": "move grasped cup to plate"}, {"step": 13, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6b9f1ba9d9e66b13306a4f11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0083/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/27", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 83, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "27", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 426, "role": "error_localization"}}
{"state": {"task": "remove beet from pot sink", "plan": [{"step": 1, "subtask": "pick up pot"}, {"step": 2, "subtask": "move grasped pot to sink"}, {"step": 3, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0084/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 84, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "put the item closest to the right burner into the cupboard", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon into the cupboard"}, {"step": 6, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0086/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 86, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 303, "role": "error_localization"}}
{"state": {"task": "Open the lip of the silver pot, move to the right upper side, then tilt it slightly to the left, then return it to its original position.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "release silver pot"}, {"step": 3, "subtask": "move grasped silver pot to the right upper side"}, {"step": 4, "subtask": "rotate grasped silver pot slightly to the left"}, {"step": 5, "subtask": "move grasped silver pot to original position"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/21e8524b6b602ef4ee0ea89b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0087/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 87, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "9", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 328, "role": "error_localization"}}
{"state": {"task": "put the silver pot on the orange cloth, then on the blue plate, then on the green tray", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to orange cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to blue plate"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to green tray"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0d5e34211222e84861c37f9d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0088/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 88, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 362, "role": "error_localization"}}
{"state": {"task": "move the green cloth to the upper right corner of the stove, then to the center burner, then to the lower left corner, then fold it neatly over the handle", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth to the upper right corner of the stove"}, {"step": 3, "subtask": "release green cloth"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth to the center burner"}, {"step": 6, "subtask": "release green cloth"}, {"step": 7, "subtask": "grasp green cloth"}, {"step": 8, "subtask": "move grasped green cloth to the lower left corner"}, {"step": 9, "subtask": "release green cloth"}, {"step": 10, "subtask": "grasp green cloth"}, {"step": 11, "subtask": "move grasped green cloth over the handle"}, {"step": 12, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13fcbb5e08966d462633de16.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0089/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 89, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the orange cloth from the right side of the table, then place it on the green towel, then move it back to the orange cloth", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the orange cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the green towel"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the orange cloth"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3928ce5b24db942b387b9579.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0090/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 90, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 377, "role": "error_localization"}}
{"state": {"task": "Move the blue spoon to the right burner, then to the left burner, then back to the right burner.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the right burner"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the left burner"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to the right burner"}, {"step": 9, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6d56a60cccc6ceb77269f32.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0091/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 91, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "Move the lid of the orange pot to the top left corner of the table, then to the bottom right, then back to the top left.", "plan": [{"step": 1, "subtask": "grasp orange pot lid"}, {"step": 2, "subtask": "move grasped orange pot lid to the top left corner of the table"}, {"step": 3, "subtask": "release orange pot lid"}, {"step": 4, "subtask": "grasp orange pot lid"}, {"step": 5, "subtask": "move grasped orange pot lid to the bottom right corner of the table"}, {"step": 6, "subtask": "release orange pot lid"}, {"step": 7, "subtask": "grasp orange pot lid"}, {"step": 8, "subtask": "move grasped orange pot lid to the top left corner of the table"}, {"step": 9, "subtask": "release orange pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1fd2ee149fa09289b75e7469.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0092/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 92, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 394, "role": "error_localization"}}
{"state": {"task": "remove the chocolate bar from the yellow cloth, then place it on the table, then return it to the cloth, then remove it again", "plan": [{"step": 1, "subtask": "grasp chocolate bar"}, {"step": 2, "subtask": "move grasped chocolate bar out of yellow cloth"}, {"step": 3, "subtask": "release chocolate bar"}, {"step": 4, "subtask": "grasp chocolate bar"}, {"step": 5, "subtask": "move grasped chocolate bar to table"}, {"step": 6, "subtask": "release chocolate bar"}, {"step": 7, "subtask": "grasp chocolate bar"}, {"step": 8, "subtask": "move grasped chocolate bar to yellow cloth"}, {"step": 9, "subtask": "release chocolate bar"}, {"step": 10, "subtask": "grasp chocolate bar"}, {"step": 11, "subtask": "move grasped chocolate bar out of yellow cloth"}, {"step": 12, "subtask": "release chocolate bar"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4afdf0776028e51c3aec77d2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0093/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_76_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 93, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_76_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 424, "role": "error_localization"}}
{"state": {"task": "Move the orange towel to the upper left side of the burner, then to the upper right side, then to the lower left side, then to the lower right side.", "plan": [{"step": 1, "subtask": "grasp orange towel"}, {"step": 2, "subtask": "move grasped orange towel to the upper left side of the burner"}, {"step": 3, "subtask": "release orange towel"}, {"step": 4, "subtask": "grasp orange towel"}, {"step": 5, "subtask": "move grasped orange towel to the upper right side of the burner"}, {"step": 6, "subtask": "release orange towel"}, {"step": 7, "subtask": "grasp orange towel"}, {"step": 8, "subtask": "move grasped orange towel to the lower left side of the burner"}, {"step": 9, "subtask": "release orange towel"}, {"step": 10, "subtask": "grasp orange towel"}, {"step": 11, "subtask": "move grasped orange towel to the lower right side of the burner"}, {"step": 12, "subtask": "release orange towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2a0442cb926d639b9efb7133.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0094/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 94, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 449, "role": "error_localization"}}
{"state": {"task": "move the pot above the left burner.", "plan": [{"step": 1, "subtask": "pick up pot"}, {"step": 2, "subtask": "move grasped pot above the left burner"}, {"step": 3, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dbc84c7a8bc41a802ad66bab.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0095/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_76_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 95, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_76_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 245, "role": "error_localization"}}
{"state": {"task": "heat the bread for 25 minutes in the oven, then put it back in the orange bowl", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp bread"}, {"step": 5, "subtask": "move grasped bread to oven"}, {"step": 6, "subtask": "release bread"}, {"step": 7, "subtask": "grasp oven door handle"}, {"step": 8, "subtask": "move grasped door handle to close position"}, {"step": 9, "subtask": "release door handle"}, {"step": 10, "subtask": "wait 25 minutes"}, {"step": 11, "subtask": "grasp door handle"}, {"step": 12, "subtask": "move grasped door handle to open position"}, {"step": 13, "subtask": "release door handle"}, {"step": 14, "subtask": "grasp orange bowl"}, {"step": 15, "subtask": "move grasped bread to orange bowl"}, {"step": 16, "subtask": "release bread"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0096/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 96, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 484, "role": "error_localization"}}
{"state": {"task": "move the red fork to the right side of the stove, then to the left side, then back to the right, then to the center", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "grasp red fork"}, {"step": 5, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 6, "subtask": "release red fork"}, {"step": 7, "subtask": "grasp red fork"}, {"step": 8, "subtask": "move grasped red fork to the center of the stove"}, {"step": 9, "subtask": "release red fork"}, {"step": 10, "subtask": "grasp red fork"}, {"step": 11, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 12, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0097/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 97, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 439, "role": "error_localization"}}
{"state": {"task": "put the pink cup and pink plate on the dryer, then put the carrot, blue cup, purple eggplant on the table", "plan": [{"step": 1, "subtask": "grasp pink cup"}, {"step": 2, "subtask": "move grasped pink cup to dryer"}, {"step": 3, "subtask": "release pink cup"}, {"step": 4, "subtask": "grasp pink plate"}, {"step": 5, "subtask": "move grasped pink plate to dryer"}, {"step": 6, "subtask": "release pink plate"}, {"step": 7, "subtask": "grasp carrot"}, {"step": 8, "subtask": "move grasped carrot to table"}, {"step": 9, "subtask": "release carrot"}, {"step": 10, "subtask": "grasp pink cup"}, {"step": 11, "subtask": "move grasped pink cup to table"}, {"step": 12, "subtask": "release pink cup"}, {"step": 13, "subtask": "grasp purple eggplant"}, {"step": 14, "subtask": "move grasped purple eggplant to table"}, {"step": 15, "subtask": "release purple eggplant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e1eedd02af5c6c03881192b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0098/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/49", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 98, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "49", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 470, "role": "error_localization"}}
{"state": {"task": "Move the silver pot lid to the top right corner of the stove, then to the top left corner, then to the bottom right corner, then to the bottom left corner.", "plan": [{"step": 1, "subtask": "grasp silver pot lid"}, {"step": 2, "subtask": "move grasped silver pot lid to the top right corner of the stove"}, {"step": 3, "subtask": "release silver pot lid"}, {"step": 4, "subtask": "grasp silver pot lid"}, {"step": 5, "subtask": "move grasped silver pot lid to the bottom left corner of the stove"}, {"step": 6, "subtask": "release silver pot lid"}, {"step": 7, "subtask": "grasp silver pot lid"}, {"step": 8, "subtask": "move grasped silver pot lid to the bottom right corner of the stove"}, {"step": 9, "subtask": "release silver pot lid"}, {"step": 10, "subtask": "grasp silver pot lid"}, {"step": 11, "subtask": "move grasped silver pot lid to the top left corner of the stove"}, {"step": 12, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6edae47522f8a97521f5736.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0099/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 99, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 462, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom of the table, then to the top, then to the middle, then to the bottom again", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom of the table"}, {"step": 3, "subtask": "grasp silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top of the table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the middle of the table"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "release silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the bottom of the table"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a628c2c636d373ca727d8859.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0100/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 100, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "12", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 434, "role": "error_localization"}}
{"state": {"task": "move the spatula on top of the orange towel, then move it to the green napkin, then back to the towel", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula to top of the orange towel"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to the green napkin"}, {"step": 6, "subtask": "release spatula"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to top of the orange towel"}, {"step": 9, "subtask": "bring it back to its original position"}, {"step": 10, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4b4a11467601363b8f736b90.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0101/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 101, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train", "episode_id": "19", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 395, "role": "error_localization"}}
{"state": {"task": "Remove the steel pot from the left burner, place it on the countertop, then move it to the sink, then back to the left burner.", "plan": [{"step": 1, "subtask": "grasp steel pot"}, {"step": 2, "subtask": "move grasped steel pot out of left burner"}, {"step": 3, "subtask": "release steel pot"}, {"step": 4, "subtask": "grasp steel pot"}, {"step": 5, "subtask": "move grasped steel pot to countertop"}, {"step": 6, "subtask": "release steel pot"}, {"step": 7, "subtask": "grasp steel pot"}, {"step": 8, "subtask": "move grasped steel pot to sink"}, {"step": 9, "subtask": "release steel pot"}, {"step": 10, "subtask": "grasp steel pot"}, {"step": 11, "subtask": "move grasped steel pot to left burner"}, {"step": 12, "subtask": "release steel pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/415ab6ceaf266d9719e4107b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0102/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_25_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 102, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_25_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 425, "role": "error_localization"}}
{"state": {"task": "remove spoon from bowl, then place it on the counter, then return it to the bowl, then place it on the sink edge", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to counter"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to bowl"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon to sink edge"}, {"step": 12, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cf10e42dde85e845f6dfc538.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0103/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/27", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 103, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "27", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "move the orange bunny to the bottom center of table, then to the top left corner, then to the bottom right corner", "plan": [{"step": 1, "subtask": "grasp orange bunny"}, {"step": 2, "subtask": "move grasped orange bunny to the bottom center of table"}, {"step": 3, "subtask": "release orange bunny"}, {"step": 4, "subtask": "grasp orange bunny"}, {"step": 5, "subtask": "move grasped orange bunny to the top left corner"}, {"step": 6, "subtask": "release orange bunny"}, {"step": 7, "subtask": "grasp orange bunny"}, {"step": 8, "subtask": "move grasped orange bunny to the bottom right corner"}, {"step": 9, "subtask": "release orange bunny"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9fe00cfd6ff3621197c89377.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0104/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/34", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 104, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "34", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 373, "role": "error_localization"}}
{"state": {"task": "Move the purple eggplant to the right side of the right burner, then to the left side of the right burner, then to the center of the right burner, then to the back side of the right burner.", "plan": [{"step": 1, "subtask": "grasp purple eggplant"}, {"step": 2, "subtask": "release purple eggplant"}, {"step": 3, "subtask": "release purple eggplant"}, {"step": 4, "subtask": "grasp purple eggplant"}, {"step": 5, "subtask": "move grasped purple eggplant to the left side of the right burner"}, {"step": 6, "subtask": "release purple eggplant"}, {"step": 7, "subtask": "grasp purple eggplant"}, {"step": 8, "subtask": "move grasped purple eggplant to the center of the right burner"}, {"step": 9, "subtask": "move grasped purple eggplant to the right side of the right burner"}, {"step": 10, "subtask": "grasp purple eggplant"}, {"step": 11, "subtask": "move grasped purple eggplant to the back side of the right burner"}, {"step": 12, "subtask": "release purple eggplant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1f17290cb276461f65ac6253.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0105/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_99_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 105, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_99_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 469, "role": "error_localization"}}
{"state": {"task": "move the red fork to the top left corner of the stove, then to the bottom right corner, then back to the top left corner", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to the top left corner of the stove"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "grasp red fork"}, {"step": 5, "subtask": "move grasped red fork to the bottom right corner of the stove"}, {"step": 6, "subtask": "release red fork"}, {"step": 7, "subtask": "grasp red fork"}, {"step": 8, "subtask": "move grasped red fork to the top left corner of the stove"}, {"step": 9, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b929de4c754f6db96d457213.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0106/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_40_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 106, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_40_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 384, "role": "error_localization"}}
{"state": {"task": "put the orange object inside the silver pot, take it out, then put it back in, then tilt the pot slightly to check stability", "plan": [{"step": 1, "subtask": "grasp orange object"}, {"step": 2, "subtask": "move grasped orange object to silver pot"}, {"step": 3, "subtask": "return it to its original position"}, {"step": 4, "subtask": "release orange object"}, {"step": 5, "subtask": "grasp orange object"}, {"step": 6, "subtask": "move grasped orange object out of silver pot"}, {"step": 7, "subtask": "release orange object"}, {"step": 8, "subtask": "grasp orange object"}, {"step": 9, "subtask": "move grasped orange object to silver pot"}, {"step": 10, "subtask": "release orange object"}, {"step": 11, "subtask": "grasp silver pot"}, {"step": 12, "subtask": "tilt grasped silver pot slightly"}, {"step": 13, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/429d4f6bfcb3d55c22a06982.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0107/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 107, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "18", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 442, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the middle right of the table, then to the front left corner, then to the back center, then return it to the middle right of the table", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the middle right of the table"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the front left corner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the back center"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the middle right of the table"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9f505a7e5472f62520bee3d2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0108/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_06_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 108, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_06_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 442, "role": "error_localization"}}
{"state": {"task": "Remove the knife from the top of the yellow cloth, place it on the table, then return it to the top of the cloth, finally move it to the cloth's left side.", "plan": [{"step": 1, "subtask": "grasp knife"}, {"step": 2, "subtask": "move grasped knife away from yellow cloth"}, {"step": 3, "subtask": "release knife"}, {"step": 4, "subtask": "grasp knife"}, {"step": 5, "subtask": "move grasped knife to top of yellow cloth"}, {"step": 6, "subtask": "release knife"}, {"step": 7, "subtask": "grasp knife"}, {"step": 8, "subtask": "move grasped knife to left side of yellow cloth"}, {"step": 9, "subtask": "release knife"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ae0bfbde3bfbca5c48f4f61.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0109/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 109, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_30_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 375, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out the spoon, place it on the table, then take out the fork and place it next to the spoon", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon out"}, {"step": 6, "subtask": "move grasped spoon to table"}, {"step": 7, "subtask": "release spoon"}, {"step": 8, "subtask": "grasp fork"}, {"step": 9, "subtask": "move grasped fork to next to spoon"}, {"step": 10, "subtask": "release fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b8d957c9dcce3f3e9dd8a509.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0110/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train/40", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 110, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train", "episode_id": "40", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 383, "role": "error_localization"}}
{"state": {"task": "moves the ribbon to left of the cloth from top, then to the right, then to the bottom, then back to the top.", "plan": [{"step": 1, "subtask": "grasp ribbon"}, {"step": 2, "subtask": "move grasped ribbon to the left of the cloth from top"}, {"step": 3, "subtask": "release ribbon"}, {"step": 4, "subtask": "grasp ribbon"}, {"step": 5, "subtask": "move grasped ribbon to the right"}, {"step": 6, "subtask": "release ribbon"}, {"step": 7, "subtask": "grasp ribbon"}, {"step": 8, "subtask": "move grasped ribbon to the bottom"}, {"step": 9, "subtask": "release ribbon"}, {"step": 10, "subtask": "grasp ribbon"}, {"step": 11, "subtask": "move grasped ribbon to the top"}, {"step": 12, "subtask": "release ribbon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4ebbc64e9b733f0257e30a7e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0111/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train/34", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 111, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_13_train", "episode_id": "34", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 415, "role": "error_localization"}}
{"state": {"task": "take the silver lid off the silver pot and put it on the left side of the silver pot", "plan": [{"step": 1, "subtask": "pick up silver lid"}, {"step": 2, "subtask": "move grasped silver lid away from silver pot"}, {"step": 3, "subtask": "release silver lid"}, {"step": 4, "subtask": "move grasped silver lid to the left side of the silver pot"}, {"step": 5, "subtask": "release silver lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e07266de493f89fdd46818e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0112/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 112, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 300, "role": "error_localization"}}
{"state": {"task": "put cucumber in cup, then take it out, then put it back in", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber to cup"}, {"step": 3, "subtask": "release cucumber"}, {"step": 4, "subtask": "grasp cucumber"}, {"step": 5, "subtask": "move grasped cucumber out"}, {"step": 6, "subtask": "release cucumber"}, {"step": 7, "subtask": "grasp cucumber"}, {"step": 8, "subtask": "move grasped cucumber to cup"}, {"step": 9, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dbce06c934d3d93ffac65ae1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0113/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/34", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 113, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "34", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out each item one by one, place them on the table, then return them one by one into the drawer", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp object"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to table"}, {"step": 7, "subtask": "release object"}, {"step": 8, "subtask": "grasp object"}, {"step": 9, "subtask": "move grasped object to drawer"}, {"step": 10, "subtask": "move grasped object backward"}, {"step": 11, "subtask": "release object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a029113e8089645841bb6d8b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0115/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/26", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 115, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "26", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 399, "role": "error_localization"}}
{"state": {"task": "put the red object on top of the orange cloth", "plan": [{"step": 1, "subtask": "move grasped red object to top of the orange cloth"}, {"step": 2, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f7f2d6ef8aa1897c31a8dd0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0116/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 116, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "put the item closest to the right burner into the cupboard", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper into the cupboard"}, {"step": 6, "subtask": "release red pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0117/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 117, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 303, "role": "error_localization"}}
{"state": {"task": "put the red pepper in the pot, then put the spatula and yellow cloth in the sink, and turn on the left burner", "plan": [{"step": 1, "subtask": "grasp red pepper"}, {"step": 2, "subtask": "move grasped red pepper to pot"}, {"step": 3, "subtask": "release red pepper"}, {"step": 4, "subtask": "grasp fork"}, {"step": 5, "subtask": "move grasped object to sink"}, {"step": 6, "subtask": "release object"}, {"step": 7, "subtask": "grasp yellow cloth"}, {"step": 8, "subtask": "move grasped yellow cloth to sink"}, {"step": 9, "subtask": "release yellow cloth"}, {"step": 10, "subtask": "grasp left burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0118/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 118, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 413, "role": "error_localization"}}
{"state": {"task": "Move the lid of the orange pot to the top left corner of the table, then to the bottom right, then back to the top left.", "plan": [{"step": 1, "subtask": "grasp orange pot lid"}, {"step": 2, "subtask": "move grasped orange pot lid to the top left corner of the table"}, {"step": 3, "subtask": "release orange pot lid"}, {"step": 4, "subtask": "grasp orange pot lid"}, {"step": 5, "subtask": "move grasped orange pot lid to the bottom right corner of the table"}, {"step": 6, "subtask": "release orange pot lid"}, {"step": 7, "subtask": "grasp orange pot lid"}, {"step": 8, "subtask": "release orange pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1fd2ee149fa09289b75e7469.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0119/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 119, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train", "episode_id": "18", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 368, "role": "error_localization"}}
{"state": {"task": "Put cup from counter to sink, then from sink to dish rack, then from dish rack to counter.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "release cup"}, {"step": 3, "subtask": "move grasped cup to sink"}, {"step": 4, "subtask": "release cup"}, {"step": 5, "subtask": "grasp cup"}, {"step": 6, "subtask": "move grasped cup to dish rack"}, {"step": 7, "subtask": "release cup"}, {"step": 8, "subtask": "grasp cup"}, {"step": 9, "subtask": "move grasped cup to counter"}, {"step": 10, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/618f735c173259d70592d0ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0120/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/43", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 120, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "43", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "Move the orange toy to the left side of the table, then to the center, then back to the left side.", "plan": [{"step": 1, "subtask": "grasp orange toy"}, {"step": 2, "subtask": "move grasped orange toy to the left side of the table"}, {"step": 3, "subtask": "release orange toy"}, {"step": 4, "subtask": "grasp orange toy"}, {"step": 5, "subtask": "move grasped orange toy to the center of the table"}, {"step": 6, "subtask": "release orange toy"}, {"step": 7, "subtask": "grasp orange toy"}, {"step": 8, "subtask": "move grasped orange toy to the left side of the table"}, {"step": 9, "subtask": "release orange toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0121/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 121, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 376, "role": "error_localization"}}
{"state": {"task": "remove the cup from the plate, then place it on the table, then move the plate to the side", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup out of plate"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp plate"}, {"step": 5, "subtask": "move grasped plate to the side"}, {"step": 6, "subtask": "release plate"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0962ac2f872e8c47534f515d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0122/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 122, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 306, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the left of burner, then to the right, then back to the left, then fold it under the burner", "plan": [{"step": 1, "subtask": "move grasped blue cloth to the left of burner"}, {"step": 2, "subtask": "release blue cloth"}, {"step": 3, "subtask": "grasp blue cloth"}, {"step": 4, "subtask": "move grasped blue cloth to the right"}, {"step": 5, "subtask": "release blue cloth"}, {"step": 6, "subtask": "move grasped blue cloth to the left of burner"}, {"step": 7, "subtask": "rotate grasped blue cloth"}, {"step": 8, "subtask": "push grasped blue cloth under burner"}, {"step": 9, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/094ab7bd16bc6b04557e54a0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0123/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 123, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 377, "role": "error_localization"}}
{"state": {"task": "Put the yellow banana in the silver pot, then remove the lid from the pot, then put the lid on the table, then remove the banana from the pot, then put the banana on the table.", "plan": [{"step": 1, "subtask": "grasp yellow banana"}, {"step": 2, "subtask": "move grasped yellow banana to silver pot"}, {"step": 3, "subtask": "release yellow banana"}, {"step": 4, "subtask": "grasp silver pot lid"}, {"step": 5, "subtask": "move grasped silver pot lid away"}, {"step": 6, "subtask": "release silver pot lid"}, {"step": 7, "subtask": "grasp yellow banana"}, {"step": 8, "subtask": "move grasped yellow banana away"}, {"step": 9, "subtask": "release yellow banana"}, {"step": 10, "subtask": "grasp yellow banana"}, {"step": 11, "subtask": "move grasped yellow banana to table"}, {"step": 12, "subtask": "release yellow banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f4c4a491a1aab38dc7ddc8db.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0124/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_36_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 124, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_36_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 434, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the left side of the burner, then to the front, then to the right side, and finally back to the left.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left side of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the front"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the right side of the burner"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the left side of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8f970e9600791f7a83997e2b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0125/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 125, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "put cup to drying rack from sink, then move it to the sink, then back to the drying rack", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to drying rack"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to sink"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to drying rack"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f862fa4cfa73b768a205f661.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0126/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 126, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "move the carrot to the drawer", "plan": [{"step": 1, "subtask": "pick up carrot"}, {"step": 2, "subtask": "open drawer"}, {"step": 3, "subtask": "move grasped carrot to drawer"}, {"step": 4, "subtask": "release carrot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc288b2bbc62dbcfc71c150c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0127/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_26_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 127, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_26_train", "episode_id": "5", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 256, "role": "error_localization"}}
{"state": {"task": "move the spoon between the burners, then place it on the stove's back burner, then move it to the front burner", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon between the burners"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to the stove's back burner"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to the front burner"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/624fcc2bfce1768bb84ae25f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0128/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 128, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 363, "role": "error_localization"}}
{"state": {"task": "place the white and red object on the edge of the cloth", "plan": [{"step": 1, "subtask": "pick up white object"}, {"step": 2, "subtask": "move grasped white object to the edge of the cloth"}, {"step": 3, "subtask": "release white object"}, {"step": 4, "subtask": "pick up red object"}, {"step": 5, "subtask": "move grasped red object to the edge of the cloth"}, {"step": 6, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/64d4c3df7464cc93a0fc4688.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0129/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_20_train/38", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 129, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_20_train", "episode_id": "38", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 309, "role": "error_localization"}}
{"state": {"task": "place the blue brush on top of the purple cloth", "plan": [{"step": 1, "subtask": "release blue brush"}, {"step": 2, "subtask": "move grasped blue brush to top of the purple cloth"}, {"step": 3, "subtask": "pick up blue brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c812d12fe8c925f72f94b05e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0130/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_79_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 130, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_79_train", "episode_id": "16", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 253, "role": "error_localization"}}
{"state": {"task": "put the cans in the cupboard, put the bread on the table, and put the red towel, and the orange bowl in the sink", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to cupboard"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp another can"}, {"step": 8, "subtask": "move grasped can to cupboard"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped bread to table"}, {"step": 12, "subtask": "release bread"}, {"step": 13, "subtask": "grasp red towel"}, {"step": 14, "subtask": "move grasped red towel to sink"}, {"step": 15, "subtask": "release red towel"}, {"step": 16, "subtask": "grasp orange"}, {"step": 17, "subtask": "move grasped orange to sink"}, {"step": 18, "subtask": "release orange"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0131/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 131, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 516, "role": "error_localization"}}
{"state": {"task": "place the blue towel at the front of the left burner", "plan": [{"step": 1, "subtask": "pick up blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the front of the left burner"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release blue towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fb63104f378cf38a3f4c526a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0132/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_88_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 132, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_88_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 274, "role": "error_localization"}}
{"state": {"task": "Remove the slice green bell pepper from the orange cloth, then place it on the table, then return it to the orange cloth", "plan": [{"step": 1, "subtask": "grasp green bell pepper"}, {"step": 2, "subtask": "move grasped green bell pepper out of the orange cloth"}, {"step": 3, "subtask": "release green bell pepper"}, {"step": 4, "subtask": "grasp green bell pepper"}, {"step": 5, "subtask": "move grasped green bell pepper to the orange cloth"}, {"step": 6, "subtask": "release green bell pepper"}, {"step": 7, "subtask": "grasp green bell pepper"}, {"step": 8, "subtask": "move grasped green bell pepper to the table"}, {"step": 9, "subtask": "release green bell pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4faf6548055fb452f6835661.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0133/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_41_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 133, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_41_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 379, "role": "error_localization"}}
{"state": {"task": "move the silver lid to the right bottom side of the table, then to the left top side, then back to the right bottom side", "plan": [{"step": 1, "subtask": "grasp silver lid"}, {"step": 2, "subtask": "move grasped silver lid to the right bottom side of the table"}, {"step": 3, "subtask": "release silver lid"}, {"step": 4, "subtask": "grasp silver lid"}, {"step": 5, "subtask": "move grasped silver lid to the left top side of the table"}, {"step": 6, "subtask": "release silver lid"}, {"step": 7, "subtask": "grasp silver lid"}, {"step": 8, "subtask": "move grasped silver lid to the right bottom side of the table"}, {"step": 9, "subtask": "release silver lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a1f5ee1edc90360a984df94d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0134/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 134, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_18_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 384, "role": "error_localization"}}
{"state": {"task": "put the tallest object to the right of the right burner", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the right of the right burner"}, {"step": 3, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0135/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 135, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 255, "role": "error_localization"}}
{"state": {"task": "move the cloth to the left bottom side of the cooker, then to the right bottom side, then fold it, then place it on the stove", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the left bottom side of the cooker"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to the right bottom side of the cooker"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to fold it"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth to the stove"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6fe0c0bb254d0d639d9420bd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0136/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 136, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 423, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl, then sink, then back to bowl, then next to sink", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to sink"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to bowl"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon next to sink"}, {"step": 12, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d690e377272db65ce808ea5d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0137/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/33", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 137, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "33", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 398, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the upper left of the table, then to the lower right, then back to the upper left, then to the center.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper left of the table"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the lower right of the table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the upper left of the table"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/411c69c79a0e1233cc485d4c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0138/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 138, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "14", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 383, "role": "error_localization"}}
{"state": {"task": "Move the spoon to the bottom right corner of the burner, then to the top left corner, then to the center of the burner, then back to the bottom right corner.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to the bottom right corner of the burner"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to the center of the burner"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to the top left corner of the burner"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon to the bottom right corner of the burner"}, {"step": 12, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3f1e7392f8ded1be8e1c0536.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0139/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 139, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 436, "role": "error_localization"}}
{"state": {"task": "take spoon out of bowl, then place spoon in sink, then take spoon out of sink, then put spoon back in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to sink"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "move grasped spoon to bowl"}, {"step": 8, "subtask": "move grasped spoon out of sink"}, {"step": 9, "subtask": "grasp spoon"}, {"step": 10, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0140/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 140, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 376, "role": "error_localization"}}
{"state": {"task": "Place the yellow cloth to the upper right side of the burner, then fold it neatly, then unfold it fully, then tuck the corners under the burner.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to the lower left side of the burner"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth to fold it neatly"}, {"step": 6, "subtask": "release yellow cloth"}, {"step": 7, "subtask": "grasp yellow cloth"}, {"step": 8, "subtask": "move grasped yellow cloth to keep it folded"}, {"step": 9, "subtask": "release yellow cloth"}, {"step": 10, "subtask": "grasp yellow cloth"}, {"step": 11, "subtask": "move grasped yellow cloth to tuck the corners under the burner"}, {"step": 12, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/78eede20c59f1c760c4da3ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0141/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_47_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 141, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_47_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 439, "role": "error_localization"}}
{"state": {"task": "Remove the slice green bell pepper from the orange cloth, then place it on the table, then return it to the orange cloth", "plan": [{"step": 1, "subtask": "grasp green bell pepper"}, {"step": 2, "subtask": "move grasped green bell pepper out of the orange cloth"}, {"step": 3, "subtask": "release green bell pepper"}, {"step": 4, "subtask": "grasp green bell pepper"}, {"step": 5, "subtask": "move grasped green bell pepper to the table"}, {"step": 6, "subtask": "release green bell pepper"}, {"step": 7, "subtask": "grasp green bell pepper"}, {"step": 8, "subtask": "move grasped green bell pepper to the orange cloth"}, {"step": 9, "subtask": "release green bell pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4faf6548055fb452f6835661.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0142/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_41_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 142, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_41_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 379, "role": "error_localization"}}
{"state": {"task": "remove the cup from the plate, then place it on the table, then pick it up again and put it back on the plate", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup out"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to plate"}, {"step": 6, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc79bd21f3dd2a49b28da0ce.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0143/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 143, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 308, "role": "error_localization"}}
{"state": {"task": "move the cloth from the left bottom side of the cooker to the top right side, then fold it, then move it to the left top side, then unfold it", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the top right side of the cooker"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "rotate grasped cloth"}, {"step": 6, "subtask": "push down"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth from right to left"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a440eaafa8ed44241d69ebf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0144/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 144, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "11", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 415, "role": "error_localization"}}
{"state": {"task": "move the cloth to the left, then to the center, then to the right, then back to the left", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "grasp cloth"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to the center"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to the right"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "move grasped cloth to the left"}, {"step": 11, "subtask": "move grasped cloth to the left"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/986e630af04cfad7046679d8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0145/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 145, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train", "episode_id": "7", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "take the blue cloth from the right side of the burner to the left, then to the front, then to the back, then fold it and place it on the counter.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth to the left"}, {"step": 3, "subtask": "move grasped blue cloth to the back"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "move grasped blue cloth to the front"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp blue cloth"}, {"step": 8, "subtask": "release blue cloth"}, {"step": 9, "subtask": "release blue cloth"}, {"step": 10, "subtask": "grasp blue cloth"}, {"step": 11, "subtask": "fold grasped blue cloth"}, {"step": 12, "subtask": "move grasped blue cloth to the counter"}, {"step": 13, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50b596b2e642d75615e7fccd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0146/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 146, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "14", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 448, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the lower side of the right burner, then to the upper side, then back to the lower side, then to the center of the stove.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower side of the right burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper side of the right burner"}, {"step": 6, "subtask": "move grasped silver pot to the lower side of the right burner"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "release silver pot"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the center of the stove"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac0da747917faa49255523b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0147/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_19_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 147, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_19_train", "episode_id": "10", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 447, "role": "error_localization"}}
{"state": {"task": "Move the yellow cloth to the left side of the silver pot, then to the right side, then back to the left side.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to the left side of the silver pot"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth to the right side of the silver pot"}, {"step": 6, "subtask": "release yellow cloth"}, {"step": 7, "subtask": "grasp yellow cloth"}, {"step": 8, "subtask": "move grasped yellow cloth to the left side of the silver pot"}, {"step": 9, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b91bfd5c180545c3532b835f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0148/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 148, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 382, "role": "error_localization"}}
{"state": {"task": "Move the green spatula from the green cloth to center of the stove, then to the front left burner, then to the front right burner, then back to the center.", "plan": [{"step": 1, "subtask": "grasp green spatula"}, {"step": 2, "subtask": "move grasped green spatula to center of the stove"}, {"step": 3, "subtask": "release green spatula"}, {"step": 4, "subtask": "grasp green spatula"}, {"step": 5, "subtask": "move grasped green spatula to front left burner"}, {"step": 6, "subtask": "release green spatula"}, {"step": 7, "subtask": "grasp green spatula"}, {"step": 8, "subtask": "move grasped green spatula to front right burner"}, {"step": 9, "subtask": "put it back where it was"}, {"step": 10, "subtask": "release green spatula"}, {"step": 11, "subtask": "grasp green spatula"}, {"step": 12, "subtask": "move grasped green spatula to center of the stove"}, {"step": 13, "subtask": "release green spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/26fc795ac507115bdda01405.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0149/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 149, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "2", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 467, "role": "error_localization"}}
{"state": {"task": "Move the blue handled spoon to the far edge of the right burner, then to the far edge of the left burner, then back to the far edge of the right burner.", "plan": [{"step": 1, "subtask": "grasp blue handled spoon"}, {"step": 2, "subtask": "move grasped blue handled spoon to the far edge of the right burner"}, {"step": 3, "subtask": "release blue handled spoon"}, {"step": 4, "subtask": "grasp blue handled spoon"}, {"step": 5, "subtask": "move grasped blue handled spoon to the far edge of the left burner"}, {"step": 6, "subtask": "release blue handled spoon"}, {"step": 7, "subtask": "grasp blue handled spoon"}, {"step": 8, "subtask": "move grasped blue handled spoon to the far edge of the right burner"}, {"step": 9, "subtask": "release blue handled spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a5ff2c67469d64cfefa024db.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0150/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_26_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 150, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_26_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 400, "role": "error_localization"}}
{"state": {"task": "Move the purple towel beside the blue cup, then under the cup, then on top of it", "plan": [{"step": 1, "subtask": "grasp purple towel"}, {"step": 2, "subtask": "move grasped purple towel beside the blue cup"}, {"step": 3, "subtask": "release purple towel"}, {"step": 4, "subtask": "grasp purple towel"}, {"step": 5, "subtask": "move grasped purple towel under the blue cup"}, {"step": 6, "subtask": "release purple towel"}, {"step": 7, "subtask": "grasp purple towel"}, {"step": 8, "subtask": "move grasped purple towel on top of the blue cup"}, {"step": 9, "subtask": "release purple towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/64864b2765e3815eec8d1db6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0151/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 151, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "Put the pot on the right upper part of the stove, then move it to the left upper part, then to the center, then back to the right upper part.", "plan": [{"step": 1, "subtask": "grasp pot"}, {"step": 2, "subtask": "move grasped pot to the right upper part of the stove"}, {"step": 3, "subtask": "release pot"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to the left upper part of the stove"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to the center of the stove"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp pot"}, {"step": 11, "subtask": "move grasped pot to the right upper part of the stove"}, {"step": 12, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/10632eb0957c40efef4f0a39.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0152/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_02_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 152, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_02_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 435, "role": "error_localization"}}
{"state": {"task": "relocate the green object to the upper right side of the drawer", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "move grasped green object to the upper right side of the drawer"}, {"step": 3, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5bb33868f20c6b1f0335082c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0154/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/23", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 154, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "23", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 256, "role": "error_localization"}}
{"state": {"task": "Place the capsicum near the ladle, then move the ladle to the center of the table, then place the capsicum next to the ladle again.", "plan": [{"step": 1, "subtask": "grasp capsicum"}, {"step": 2, "subtask": "move grasped capsicum near the ladle"}, {"step": 3, "subtask": "release capsicum"}, {"step": 4, "subtask": "grasp ladle"}, {"step": 5, "subtask": "move grasped ladle to the center of the table"}, {"step": 6, "subtask": "release ladle"}, {"step": 7, "subtask": "grasp capsicum"}, {"step": 8, "subtask": "move grasped capsicum next to the ladle"}, {"step": 9, "subtask": "release capsicum"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/96837ab6c4383dc63e004864.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0155/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_104_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 155, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_104_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 380, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the can that is the closest to the cyan towel", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp yellow can"}, {"step": 11, "subtask": "move grasped yellow can to upper right corner"}, {"step": 12, "subtask": "release yellow can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0156/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 156, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "Move the cloth to the top left of the table, then fold it, then unfold it.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top left of the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth forward"}, {"step": 6, "subtask": "move grasped cloth backward"}, {"step": 7, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6dda6324fc896d5e53761eb4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0157/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 157, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 321, "role": "error_localization"}}
{"state": {"task": "put the cloth on the table to the burner", "plan": [{"step": 1, "subtask": "pick up cloth"}, {"step": 2, "subtask": "move grasped cloth to the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "pick up cloth"}, {"step": 5, "subtask": "move grasped cloth to the burner"}, {"step": 6, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6289f95cd3b7f3e99705c663.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0158/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 158, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 294, "role": "error_localization"}}
{"state": {"task": "move the green object in the pot above the right burner.", "plan": [{"step": 1, "subtask": "pick up green object"}, {"step": 2, "subtask": "move grasped green object to pot"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "pick up pot"}, {"step": 5, "subtask": "move grasped pot above the right burner"}, {"step": 6, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd2bd32f33d4e6aa80411a08.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0159/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_86_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 159, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_86_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 299, "role": "error_localization"}}
{"state": {"task": "Put the silver pot on the blue cloth, then move it to the yellow cloth, then back to the blue cloth.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to blue cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to yellow cloth"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to blue cloth"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/19e55cd71a6ac7eab8da495e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0160/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 160, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 365, "role": "error_localization"}}
{"state": {"task": "Put up the silver pot on the right burner, then move it to the left burner, then return it to the right burner.", "plan": [{"step": 1, "subtask": "move grasped silver pot to the left burner"}, {"step": 2, "subtask": "release silver pot"}, {"step": 3, "subtask": "move grasped silver pot to the right burner"}, {"step": 4, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14c5a0d430c195739fc480ee.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0162/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_06_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 162, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_06_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 286, "role": "error_localization"}}
{"state": {"task": "move the green cloth to the upper right corner of the stove, then to the center burner, then to the lower left corner, then fold it neatly over the handle", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth to the upper right corner of the stove"}, {"step": 3, "subtask": "release green cloth"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth to the lower left corner"}, {"step": 6, "subtask": "release green cloth"}, {"step": 7, "subtask": "grasp green cloth"}, {"step": 8, "subtask": "move grasped green cloth to the center burner"}, {"step": 9, "subtask": "release green cloth"}, {"step": 10, "subtask": "grasp green cloth"}, {"step": 11, "subtask": "move grasped green cloth over the handle"}, {"step": 12, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13fcbb5e08966d462633de16.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0163/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 163, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "Put the green cloth on the right side of the burner, then fold it neatly, then unfold it back.", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth to the right side of the burner"}, {"step": 3, "subtask": "release green cloth"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth forward"}, {"step": 6, "subtask": "move grasped green cloth backward"}, {"step": 7, "subtask": "release green cloth"}, {"step": 8, "subtask": "grasp green cloth"}, {"step": 9, "subtask": "move grasped green cloth from right to left"}, {"step": 10, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5059793a7956e057cde276d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0164/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 164, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "4", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "take the red object out of the drawer and put it in front", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up red object"}, {"step": 3, "subtask": "move grasped red object to front"}, {"step": 4, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8f1310fdac2925e3d3086af7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0165/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train/41", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 165, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train", "episode_id": "41", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 266, "role": "error_localization"}}
{"state": {"task": "Move the napkin to the left side of the blue fork, then move it to the right side of the red plate, then move it under the green cup.", "plan": [{"step": 1, "subtask": "grasp napkin"}, {"step": 2, "subtask": "move grasped napkin to the left side of the blue fork"}, {"step": 3, "subtask": "release napkin"}, {"step": 4, "subtask": "grasp napkin"}, {"step": 5, "subtask": "move grasped napkin to the right side of the red plate"}, {"step": 6, "subtask": "release napkin"}, {"step": 7, "subtask": "grasp napkin"}, {"step": 8, "subtask": "move grasped napkin under the green cup"}, {"step": 9, "subtask": "release napkin"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/98d58ec9799f25aac67d0e37.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0166/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 166, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "Take the spoon out of the bowl, sink, then place it on the table, then put it back in the bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "move grasped spoon to sink"}, {"step": 4, "subtask": "release spoon"}, {"step": 5, "subtask": "grasp spoon"}, {"step": 6, "subtask": "move grasped spoon to table"}, {"step": 7, "subtask": "release spoon"}, {"step": 8, "subtask": "grasp spoon"}, {"step": 9, "subtask": "move grasped spoon to bowl"}, {"step": 10, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d87578ae3b0f18b93da42ac.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0167/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/20", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 167, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 374, "role": "error_localization"}}
{"state": {"task": "Put the item closest to the left burner into the pot", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped object to pot"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0168/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 168, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 247, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out the apple, place it on the table, then return it to the drawer, close the drawer.", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp apple"}, {"step": 5, "subtask": "move grasped apple out"}, {"step": 6, "subtask": "move grasped apple to floor"}, {"step": 7, "subtask": "release apple"}, {"step": 8, "subtask": "grasp apple"}, {"step": 9, "subtask": "move grasped apple backward"}, {"step": 10, "subtask": "move grasped apple to drawer"}, {"step": 11, "subtask": "release apple"}, {"step": 12, "subtask": "grasp drawer handle"}, {"step": 13, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 14, "subtask": "release drawer handle"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9fbbca24a8a9fbb667d1f19c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0169/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 169, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 453, "role": "error_localization"}}
{"state": {"task": "put the yellow corn in the pot, put the pot on the right burner and turn it on", "plan": [{"step": 1, "subtask": "grasp yellow corn"}, {"step": 2, "subtask": "move grasped yellow corn to the pot"}, {"step": 3, "subtask": "release yellow corn"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to the right burner"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp left burner knob"}, {"step": 8, "subtask": "rotate grasped object to on position"}, {"step": 9, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac62285427547839b7682924.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0170/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 170, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "Move the blue object to the top right spot of the right burner, then to the center, then to the bottom left spot, then back to the top right spot.", "plan": [{"step": 1, "subtask": "grasp blue object"}, {"step": 2, "subtask": "move grasped blue object to the top right spot of the right burner"}, {"step": 3, "subtask": "release blue object"}, {"step": 4, "subtask": "grasp blue object"}, {"step": 5, "subtask": "move grasped blue object to the center"}, {"step": 6, "subtask": "release blue object"}, {"step": 7, "subtask": "grasp blue object"}, {"step": 8, "subtask": "move grasped blue object to the bottom left spot"}, {"step": 9, "subtask": "release blue object"}, {"step": 10, "subtask": "grasp blue object"}, {"step": 11, "subtask": "move grasped blue object to the top right spot of the right burner"}, {"step": 12, "subtask": "release blue object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4abc430ddbc71789b2a7473a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0171/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_46_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 171, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_46_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 443, "role": "error_localization"}}
{"state": {"task": "Put the silver pot to the left bottom of the burner, then move it to the top center, and finally return it to the left bottom of the burner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left bottom of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top center of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the left bottom of the burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/80ede11340938c56665a58e6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0172/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 172, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "put into the cupboard the pot, spatula, red pepper, yellow towel, remove from the cupboard a fork and spatula and put them to the table", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to cupboard"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to cupboard"}, {"step": 9, "subtask": "release spatula"}, {"step": 10, "subtask": "grasp red pepper"}, {"step": 11, "subtask": "move grasped red pepper to cupboard"}, {"step": 12, "subtask": "release red pepper"}, {"step": 13, "subtask": "grasp yellow towel"}, {"step": 14, "subtask": "move grasped yellow towel to cupboard"}, {"step": 15, "subtask": "release yellow towel"}, {"step": 16, "subtask": "grasp fork"}, {"step": 17, "subtask": "move grasped fork out of cupboard"}, {"step": 18, "subtask": "move grasped fork to table"}, {"step": 19, "subtask": "release fork"}, {"step": 20, "subtask": "grasp spatula"}, {"step": 21, "subtask": "move grasped spatula out of cupboard"}, {"step": 22, "subtask": "move grasped spatula to table"}, {"step": 23, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0173/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 173, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 619, "role": "error_localization"}}
{"state": {"task": "Move the yellow object to the pot from the left of the left side burner, then return it to its original position, then move it to the right of the right side burner.", "plan": [{"step": 1, "subtask": "grasp yellow object"}, {"step": 2, "subtask": "move grasped yellow object to the pot from the left of the left side burner"}, {"step": 3, "subtask": "release yellow object"}, {"step": 4, "subtask": "grasp yellow object"}, {"step": 5, "subtask": "move grasped yellow object to original position"}, {"step": 6, "subtask": "release yellow object"}, {"step": 7, "subtask": "grasp yellow object"}, {"step": 8, "subtask": "move grasped yellow object to the right of the right side burner"}, {"step": 9, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8eff56aaacb09f759a791e04.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0174/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 174, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 390, "role": "error_localization"}}
{"state": {"task": "Move the red pepper on top of the orange cloth, then on the blue plate, then back to the orange cloth.", "plan": [{"step": 1, "subtask": "grasp red pepper"}, {"step": 2, "subtask": "move grasped red pepper to top of the orange cloth"}, {"step": 3, "subtask": "release red pepper"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to blue plate"}, {"step": 6, "subtask": "release red pepper"}, {"step": 7, "subtask": "grasp red pepper"}, {"step": 8, "subtask": "move grasped red pepper to top of the orange cloth"}, {"step": 9, "subtask": "release red pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7461e053a5b6e69086656776.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0175/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 175, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the right bottom of the burner, then to the left top, then to the center, then back to the right bottom", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the right bottom of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the left top of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center of the burner"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the right bottom of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5c33dae017f4a15ba5e162f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0176/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_56_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 176, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_56_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 440, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out the pen, place it on the table, then take out the book, place it next to the pen, then take out the key and place it on top of the book", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp pen"}, {"step": 5, "subtask": "move grasped pen out"}, {"step": 6, "subtask": "move grasped pen to drawer"}, {"step": 7, "subtask": "release pen"}, {"step": 8, "subtask": "grasp book"}, {"step": 9, "subtask": "move grasped book out"}, {"step": 10, "subtask": "move grasped book to next to pen"}, {"step": 11, "subtask": "release book"}, {"step": 12, "subtask": "grasp key"}, {"step": 13, "subtask": "move grasped key out"}, {"step": 14, "subtask": "move grasped key to top of book"}, {"step": 15, "subtask": "release key"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a46a657d90cfb88bc5dbc5ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0177/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 177, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 485, "role": "error_localization"}}
{"state": {"task": "put the item that is between the sink and stove into the cupboard", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon into the cupboard"}, {"step": 6, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0178/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 178, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 305, "role": "error_localization"}}
{"state": {"task": "move the blue fork to the upper left side of the stove next to the red object, then to the lower right side, then back to the upper left side", "plan": [{"step": 1, "subtask": "grasp blue fork"}, {"step": 2, "subtask": "move grasped blue fork to the upper left side of the stove next to the red object"}, {"step": 3, "subtask": "release blue fork"}, {"step": 4, "subtask": "grasp blue fork"}, {"step": 5, "subtask": "move grasped blue fork to the lower right side"}, {"step": 6, "subtask": "release blue fork"}, {"step": 7, "subtask": "grasp blue fork"}, {"step": 8, "subtask": "move grasped blue fork to the upper left side of the stove next to the red object"}, {"step": 9, "subtask": "release blue fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d361addd4233dde18166a4f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0179/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_107_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 179, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_107_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 396, "role": "error_localization"}}
{"state": {"task": "Put the item closest to the sink into the sink", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to sink"}, {"step": 3, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0180/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 180, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 249, "role": "error_localization"}}
{"state": {"task": "Put cup to counter from sink, then to the stove, then back to the counter, then to the sink.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to counter"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to sink"}, {"step": 12, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6dc268665d15813c6f2c7446.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0181/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 181, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 402, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the top left corner of the cook from the left burner, then to the top right corner, then to the bottom left corner, then back to the left burner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the top left corner of the cook from the left burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top right corner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the bottom left corner"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the left burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aca3a0b630d2b7fd14bfef3e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0182/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_81_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 182, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_81_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 447, "role": "error_localization"}}
{"state": {"task": "take spatula off plate sink, then place it on the counter, then pick it up again, then return it to the plate sink", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula out"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "release spatula"}, {"step": 6, "subtask": "move grasped spatula to counter"}, {"step": 7, "subtask": "release spatula"}, {"step": 8, "subtask": "grasp spatula"}, {"step": 9, "subtask": "move grasped spatula to plate sink"}, {"step": 10, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/09c16a5ebc339db4921d2890.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0183/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spatula_off_plate_sink_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 183, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spatula_off_plate_sink_train", "episode_id": "21", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 382, "role": "error_localization"}}
{"state": {"task": "cook the red pepper in the pot with the oven for 5 minutes, put the pot on the stove, close the oven, then remove the red pepper from the pot, put it on the table, put the pot in the sink", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to pot"}, {"step": 6, "subtask": "release red pepper"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to oven"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp oven door handle"}, {"step": 11, "subtask": "move grasped door handle to close position"}, {"step": 12, "subtask": "release door handle"}, {"step": 13, "subtask": "wait 5 minutes"}, {"step": 14, "subtask": "grasp oven door handle"}, {"step": 15, "subtask": "move grasped door handle to open position"}, {"step": 16, "subtask": "release door handle"}, {"step": 17, "subtask": "grasp pot"}, {"step": 18, "subtask": "move grasped pot to stove"}, {"step": 19, "subtask": "release pot"}, {"step": 20, "subtask": "grasp oven door handle"}, {"step": 21, "subtask": "move grasped door handle to close position"}, {"step": 22, "subtask": "release door handle"}, {"step": 23, "subtask": "grasp red pepper"}, {"step": 24, "subtask": "move grasped red pepper to table"}, {"step": 25, "subtask": "release red pepper"}, {"step": 26, "subtask": "grasp spatula"}, {"step": 27, "subtask": "move grasped spatula to sink"}, {"step": 28, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0184/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 184, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 723, "role": "error_localization"}}
{"state": {"task": "put the blue cloth on the top edge of the cloth, then move it to the bottom edge, then return it to the top edge.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth to the top edge of the cloth"}, {"step": 3, "subtask": "release blue cloth"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "move grasped blue cloth to the bottom edge of the cloth"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp blue cloth"}, {"step": 8, "subtask": "move grasped blue cloth to the top edge of the cloth"}, {"step": 9, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e9580694af303987d601bcb5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0185/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 185, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_14_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 381, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the upper left side of the table, then to the upper right side, then to the lower left side, then to the center of the table", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper left side of the table"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper right side of the table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the lower left side of the table"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the center of the table"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5691cd8f1cb9f6f6d3f3eaf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0186/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 186, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 448, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the can that is the closest to the cyan towel", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cyan towel"}, {"step": 11, "subtask": "move grasped cyan towel to upper right corner"}, {"step": 12, "subtask": "release cyan towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0187/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 187, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "place the wok on the purple napkin, then move it to the stove, then back to the napkin, then next to the napkin.", "plan": [{"step": 1, "subtask": "grasp wok"}, {"step": 2, "subtask": "move grasped wok to purple napkin"}, {"step": 3, "subtask": "release wok"}, {"step": 4, "subtask": "grasp wok"}, {"step": 5, "subtask": "move grasped wok to stove"}, {"step": 6, "subtask": "grasp wok"}, {"step": 7, "subtask": "grasp wok"}, {"step": 8, "subtask": "move grasped wok to purple napkin"}, {"step": 9, "subtask": "release wok"}, {"step": 10, "subtask": "release wok"}, {"step": 11, "subtask": "move grasped wok next to purple napkin"}, {"step": 12, "subtask": "release wok"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7f91aada8a55f5a3afb809ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0188/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_101_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 188, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_101_train", "episode_id": "17", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 416, "role": "error_localization"}}
{"state": {"task": "take spoon out of bowl", "plan": [{"step": 1, "subtask": "pick up spoon"}, {"step": 2, "subtask": "move grasped spoon to sink"}, {"step": 3, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/48b2ee0f1c49b1bc4c6e81b0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0189/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/28", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 189, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "28", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "Move the yellow brush to the upper left corner of the stove, then to the upper right corner, then to the lower left corner, then back to the upper left corner.", "plan": [{"step": 1, "subtask": "grasp yellow brush"}, {"step": 2, "subtask": "move grasped yellow brush to the upper left corner of the stove"}, {"step": 3, "subtask": "release yellow brush"}, {"step": 4, "subtask": "grasp yellow brush"}, {"step": 5, "subtask": "move grasped yellow brush to the upper right corner of the stove"}, {"step": 6, "subtask": "release yellow brush"}, {"step": 7, "subtask": "grasp yellow brush"}, {"step": 8, "subtask": "move grasped yellow brush to the lower left corner of the stove"}, {"step": 9, "subtask": "release yellow brush"}, {"step": 10, "subtask": "grasp yellow brush"}, {"step": 11, "subtask": "move grasped yellow brush to the upper left corner of the stove"}, {"step": 12, "subtask": "release yellow brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c480b856b155bdef3d96843.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0190/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_93_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 190, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_93_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 450, "role": "error_localization"}}
{"state": {"task": "Remove the cucumber from the cup, then place the cup upside down, then return the cucumber to the cup.", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber out of cup"}, {"step": 3, "subtask": "release cucumber"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "rotate grasped cup"}, {"step": 6, "subtask": "move grasped cup to upside down position"}, {"step": 7, "subtask": "release cup"}, {"step": 8, "subtask": "grasp cucumber"}, {"step": 9, "subtask": "move grasped cucumber to cup"}, {"step": 10, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e16132c4362e37544584ee45.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0191/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/36", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 191, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "36", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "move the slice of orange on the side of the table, then move it to the center, then move it back to the side, then pick up another slice and place it next to the first", "plan": [{"step": 1, "subtask": "grasp orange slice"}, {"step": 2, "subtask": "move grasped orange slice to the side of the table"}, {"step": 3, "subtask": "release orange slice"}, {"step": 4, "subtask": "grasp orange slice"}, {"step": 5, "subtask": "move grasped orange slice to the center of the table"}, {"step": 6, "subtask": "release orange slice"}, {"step": 7, "subtask": "grasp orange slice"}, {"step": 8, "subtask": "move grasped orange slice to the side of the table"}, {"step": 9, "subtask": "release orange slice"}, {"step": 10, "subtask": "grasp another orange slice"}, {"step": 11, "subtask": "move grasped orange slice next to the first orange slice"}, {"step": 12, "subtask": "release orange slice"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/082fb7b21b913ced4428ff74.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0192/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 192, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 448, "role": "error_localization"}}
{"state": {"task": "Remove blueberries from plate, place them on counter, then return them to plate, then move them to sink again", "plan": [{"step": 1, "subtask": "release blueberries"}, {"step": 2, "subtask": "move grasped blueberries to counter"}, {"step": 3, "subtask": "release blueberries"}, {"step": 4, "subtask": "grasp blueberries"}, {"step": 5, "subtask": "move grasped blueberries to plate"}, {"step": 6, "subtask": "grasp blueberries"}, {"step": 7, "subtask": "grasp blueberries"}, {"step": 8, "subtask": "move grasped blueberries to sink"}, {"step": 9, "subtask": "release blueberries"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/982439ed1a963a94e8513cfe.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0193/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 193, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "1", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 362, "role": "error_localization"}}
{"state": {"task": "Move the green object into the silver pot, then out of the pot, then place it beside the pot, then back into the pot.", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object to silver pot"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object out of silver pot"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "grasp green object"}, {"step": 8, "subtask": "move grasped green object on top of silver pot"}, {"step": 9, "subtask": "release green object"}, {"step": 10, "subtask": "grasp green object"}, {"step": 11, "subtask": "move grasped green object to silver pot"}, {"step": 12, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0194/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 194, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 426, "role": "error_localization"}}
{"state": {"task": "open the oven, then move the pot to the open oven door, then close the oven, then move the pan to the open oven door", "plan": [{"step": 1, "subtask": "grasp oven handle"}, {"step": 2, "subtask": "move grasped oven handle to oven's open position"}, {"step": 3, "subtask": "release oven handle"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to open oven door"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp oven handle"}, {"step": 8, "subtask": "move grasped oven handle to oven's closed position"}, {"step": 9, "subtask": "release oven handle"}, {"step": 10, "subtask": "move grasped pan to open oven door"}, {"step": 11, "subtask": "release pan"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3b82692e6e4e5950f12ad350.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0195/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_close_oven_train/38", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 195, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_close_oven_train", "episode_id": "38", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 408, "role": "error_localization"}}
{"state": {"task": "cook the red pepper in the pot with the oven for 5 minutes, put the pot on the stove, close the oven, then remove the red pepper from the pot, put it on the table, put the pot in the sink", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "move grasped red apple to pot"}, {"step": 6, "subtask": "release red apple"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to oven"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp oven door handle"}, {"step": 11, "subtask": "move grasped door handle to close position"}, {"step": 12, "subtask": "release door handle"}, {"step": 13, "subtask": "wait 5 minutes"}, {"step": 14, "subtask": "grasp oven door handle"}, {"step": 15, "subtask": "move grasped door handle to open position"}, {"step": 16, "subtask": "release door handle"}, {"step": 17, "subtask": "grasp pot"}, {"step": 18, "subtask": "move grasped pot to stove"}, {"step": 19, "subtask": "release pot"}, {"step": 20, "subtask": "grasp oven door handle"}, {"step": 21, "subtask": "move grasped door handle to close position"}, {"step": 22, "subtask": "release door handle"}, {"step": 23, "subtask": "grasp red pepper"}, {"step": 24, "subtask": "move grasped red pepper to table"}, {"step": 25, "subtask": "release red pepper"}, {"step": 26, "subtask": "grasp pot"}, {"step": 27, "subtask": "move grasped pot to sink"}, {"step": 28, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0196/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 196, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 720, "role": "error_localization"}}
{"state": {"task": "remove the yellow duck from the drawer", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up yellow duck"}, {"step": 3, "subtask": "move grasped yellow duck away from drawer"}, {"step": 4, "subtask": "release yellow duck"}, {"step": 5, "subtask": "close drawer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cadf3573d63e2515c5e0acd8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0197/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/30", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 197, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "30", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 275, "role": "error_localization"}}
{"state": {"task": "place the silver bowl on top of the yellow towel on the left side of the stove", "plan": [{"step": 1, "subtask": "pick up yellow towel"}, {"step": 2, "subtask": "move grasped yellow towel to the left side of the stove"}, {"step": 3, "subtask": "release yellow towel"}, {"step": 4, "subtask": "pick up silver bowl"}, {"step": 5, "subtask": "move grasped silver bowl to the top of the yellow towel"}, {"step": 6, "subtask": "release silver bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ce43368a343061d9032ce8b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0198/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_09_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 198, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_09_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 316, "role": "error_localization"}}
{"state": {"task": "put cucumber in cup, then take it out, then put it back in", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber to cup"}, {"step": 3, "subtask": "move grasped cucumber to cup"}, {"step": 4, "subtask": "grasp cucumber"}, {"step": 5, "subtask": "move grasped cucumber out"}, {"step": 6, "subtask": "release cucumber"}, {"step": 7, "subtask": "grasp cucumber"}, {"step": 8, "subtask": "release cucumber"}, {"step": 9, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dbce06c934d3d93ffac65ae1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0199/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/34", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 199, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "34", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "Move the green toy to the bottom of the table, then to the top, then to the left side, then back to the bottom.", "plan": [{"step": 1, "subtask": "grasp green toy"}, {"step": 2, "subtask": "move grasped green toy to the bottom of the table"}, {"step": 3, "subtask": "release green toy"}, {"step": 4, "subtask": "grasp green toy"}, {"step": 5, "subtask": "move grasped green toy to the top of the table"}, {"step": 6, "subtask": "release green toy"}, {"step": 7, "subtask": "grasp green toy"}, {"step": 8, "subtask": "release green toy"}, {"step": 9, "subtask": "move grasped green toy to the bottom of the table"}, {"step": 10, "subtask": "release green toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b98ea03009591c137394fc15.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0200/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 200, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train", "episode_id": "4", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 394, "role": "error_localization"}}
{"state": {"task": "move the spatula on top of the orange towel, then move it to the green napkin, then back to the towel", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula to top of the orange towel"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to the green napkin"}, {"step": 6, "subtask": "release spatula"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to top of the orange towel"}, {"step": 9, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4b4a11467601363b8f736b90.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0201/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 201, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 375, "role": "error_localization"}}
{"state": {"task": "put the item that is between the sink and stove into the cupboard", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper into the cupboard"}, {"step": 6, "subtask": "release red pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0202/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 202, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 305, "role": "error_localization"}}
{"state": {"task": "move the cloth to right from left, then fold it, then unfold it", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the right"}, {"step": 3, "subtask": "bring it back to its original position"}, {"step": 4, "subtask": "release cloth"}, {"step": 5, "subtask": "grasp cloth"}, {"step": 6, "subtask": "fold cloth"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "unfold cloth"}, {"step": 10, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0203/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 203, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "move the orange cloth to the left burner", "plan": [{"step": 1, "subtask": "move grasped orange cloth to the left burner"}, {"step": 2, "subtask": "pick up orange cloth"}, {"step": 3, "subtask": "release orange cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7bbd5336fef654e1c6796542.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0204/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_02_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 204, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_02_train", "episode_id": "3", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 249, "role": "error_localization"}}
{"state": {"task": "heat the bread for 25 minutes in the oven, then put it back in the orange bowl", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp bread"}, {"step": 5, "subtask": "move grasped bread to oven"}, {"step": 6, "subtask": "release bread"}, {"step": 7, "subtask": "grasp oven door handle"}, {"step": 8, "subtask": "move grasped door handle to close position"}, {"step": 9, "subtask": "release door handle"}, {"step": 10, "subtask": "wait 25 minutes"}, {"step": 11, "subtask": "grasp bowl handle"}, {"step": 12, "subtask": "move grasped door handle to open position"}, {"step": 13, "subtask": "release door handle"}, {"step": 14, "subtask": "grasp bread"}, {"step": 15, "subtask": "move grasped bread to orange bowl"}, {"step": 16, "subtask": "release bread"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0205/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 205, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 483, "role": "error_localization"}}
{"state": {"task": "remove the chocolate bar from the yellow cloth, then place it on the table, then return it to the cloth, then remove it again", "plan": [{"step": 1, "subtask": "grasp chocolate bar"}, {"step": 2, "subtask": "move grasped chocolate bar out of yellow cloth"}, {"step": 3, "subtask": "release chocolate bar"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth to table"}, {"step": 6, "subtask": "release yellow cloth"}, {"step": 7, "subtask": "grasp chocolate bar"}, {"step": 8, "subtask": "move grasped chocolate bar to yellow cloth"}, {"step": 9, "subtask": "release chocolate bar"}, {"step": 10, "subtask": "grasp chocolate bar"}, {"step": 11, "subtask": "move grasped chocolate bar out of yellow cloth"}, {"step": 12, "subtask": "release chocolate bar"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4afdf0776028e51c3aec77d2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0206/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_76_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 206, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_76_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 424, "role": "error_localization"}}
{"state": {"task": "move the blue spatula on the left burner, then move it to the right burner, then move it back to the left burner", "plan": [{"step": 1, "subtask": "grasp blue spatula"}, {"step": 2, "subtask": "move grasped blue spatula to middle burner"}, {"step": 3, "subtask": "release blue spatula"}, {"step": 4, "subtask": "grasp blue spatula"}, {"step": 5, "subtask": "move grasped blue spatula to right burner"}, {"step": 6, "subtask": "release blue spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/783b3ec5d8c46e163ee8e7c4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0207/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 207, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 323, "role": "error_localization"}}
{"state": {"task": "move the cloth from the bottom right to the top left of the table, then fold it in half, then move it back to the bottom right", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the top left of the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold grasped cloth in half"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "move grasped cloth to the bottom right of the table"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf568029e1d1a3734b86d995.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0208/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 208, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train", "episode_id": "17", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 369, "role": "error_localization"}}
{"state": {"task": "take spoon out of bowl, then place it on the dish rack, then move it to the counter", "plan": [{"step": 1, "subtask": "move grasped spoon to dish rack"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "grasp spoon"}, {"step": 4, "subtask": "release spoon"}, {"step": 5, "subtask": "grasp spoon"}, {"step": 6, "subtask": "move grasped spoon to counter"}, {"step": 7, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8910d4c69f20807a610ef6e3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0209/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 209, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "18", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 323, "role": "error_localization"}}
{"state": {"task": "Place the silver pot on the yellow cloth, then move it to the red plate, then back to the yellow cloth.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to yellow cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to red plate"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to yellow cloth"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e1914225b7850e380f1b5ff0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0211/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_25_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 211, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_25_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 365, "role": "error_localization"}}
{"state": {"task": "move the cloth to the bottom right, then to the top left, then fold it, then to the center", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the bottom right"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "release cloth"}, {"step": 6, "subtask": "move grasped cloth to the top left"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "rotate grasped cloth"}, {"step": 10, "subtask": "push down"}, {"step": 11, "subtask": "release cloth"}, {"step": 12, "subtask": "grasp cloth"}, {"step": 13, "subtask": "move grasped cloth to the center"}, {"step": 14, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/67c7b1f2958e652cec2c5c34.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0212/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 212, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train", "episode_id": "17", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 435, "role": "error_localization"}}
{"state": {"task": "put the cans in the cupboard, put the bread on the table, and put the red towel, and the orange bowl in the sink", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to cupboard"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to cupboard"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped bread to table"}, {"step": 12, "subtask": "release bread"}, {"step": 13, "subtask": "grasp red towel"}, {"step": 14, "subtask": "move grasped red towel to sink"}, {"step": 15, "subtask": "release red towel"}, {"step": 16, "subtask": "grasp orange bowl"}, {"step": 17, "subtask": "move grasped orange bowl to sink"}, {"step": 18, "subtask": "release orange bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0213/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 213, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 521, "role": "error_localization"}}
{"state": {"task": "take the silver pot lid, place it on the silver pot, then remove it, then place it again", "plan": [{"step": 1, "subtask": "grasp silver pot lid"}, {"step": 2, "subtask": "move grasped silver pot lid to silver pot"}, {"step": 3, "subtask": "release silver pot lid"}, {"step": 4, "subtask": "move grasped silver pot lid to silver pot"}, {"step": 5, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8f732e0d674b04d9d22289f8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0214/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 214, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "0", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 301, "role": "error_localization"}}
{"state": {"task": "place the yellow brush on the blue cloth, then move the blue cloth to the green tray, then place the yellow brush back on the blue cloth.", "plan": [{"step": 1, "subtask": "grasp yellow brush"}, {"step": 2, "subtask": "move grasped yellow brush to blue cloth"}, {"step": 3, "subtask": "release yellow brush"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "release blue cloth"}, {"step": 6, "subtask": "grasp yellow brush"}, {"step": 7, "subtask": "move grasped yellow brush to blue cloth"}, {"step": 8, "subtask": "release yellow brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0ae8aa96ce4023993413e5e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0215/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_93_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 215, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_93_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "pick up blue pen and put into drawer", "plan": [{"step": 1, "subtask": "pick up drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "pick up blue pen"}, {"step": 5, "subtask": "move grasped blue pen to drawer"}, {"step": 6, "subtask": "release blue pen"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3c336d1d9cc1fcf9c8f1daf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0216/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/23", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 216, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "23", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 300, "role": "error_localization"}}
{"state": {"task": "move the lemon on the right side of the stove, then to the left side, then back to the right side, then to the front side of the stove", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to the right side of the stove"}, {"step": 3, "subtask": "release lemon"}, {"step": 4, "subtask": "grasp lemon"}, {"step": 5, "subtask": "move grasped lemon to the right side of the stove"}, {"step": 6, "subtask": "release lemon"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped lemon to the front side of the stove"}, {"step": 9, "subtask": "release lemon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6298e643302c41acc33c2f7d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0217/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_24_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 217, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_24_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 377, "role": "error_localization"}}
{"state": {"task": "put the yellow towel at the bottom of the left burner", "plan": [{"step": 1, "subtask": "pick up yellow towel"}, {"step": 2, "subtask": "move grasped yellow towel to the bottom of the left burner"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release yellow towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3ade824fdf797412725045a9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0218/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 218, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "12", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 274, "role": "error_localization"}}
{"state": {"task": "Put cup to drying rack from sink, then move plate from sink to drying rack, then fork from sink to drying rack", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "release cup"}, {"step": 3, "subtask": "grasp plate"}, {"step": 4, "subtask": "move grasped plate to drying rack"}, {"step": 5, "subtask": "release plate"}, {"step": 6, "subtask": "move grasped fork to drying rack"}, {"step": 7, "subtask": "release fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f0350fc1da496a4bb9b0fc6d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0219/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 219, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "5", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 323, "role": "error_localization"}}
{"state": {"task": "Put the bread on top of the purple cloth, then move it to the right, then to the left, then back to the center of the cloth", "plan": [{"step": 1, "subtask": "grasp bread"}, {"step": 2, "subtask": "move grasped bread to top of the purple cloth"}, {"step": 3, "subtask": "release bread"}, {"step": 4, "subtask": "grasp bread"}, {"step": 5, "subtask": "move grasped bread to the right"}, {"step": 6, "subtask": "release bread"}, {"step": 7, "subtask": "grasp bread"}, {"step": 8, "subtask": "move grasped bread to the left"}, {"step": 9, "subtask": "release bread"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped bread to the center of the cloth"}, {"step": 12, "subtask": "release bread"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2e17dbfc12584ead392a5623.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0220/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 220, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 420, "role": "error_localization"}}
{"state": {"task": "Take the blue cloth to the right side of the burner from the left, then to the left side, then back to the right.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth to the right side of the burner"}, {"step": 3, "subtask": "release blue cloth"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "move grasped blue cloth to the left side of the burner"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp blue cloth"}, {"step": 8, "subtask": "move grasped blue cloth to the right side of the burner"}, {"step": 9, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1220b0cd25acbee2da4bebc8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0221/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 221, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 380, "role": "error_localization"}}
{"state": {"task": "place the eggplant on top of the green cloth, then move the cloth with the eggplant to the table, then remove the eggplant and place it on the floor", "plan": [{"step": 1, "subtask": "grasp eggplant"}, {"step": 2, "subtask": "move grasped eggplant to top of the green cloth"}, {"step": 3, "subtask": "release eggplant"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth to the floor"}, {"step": 6, "subtask": "release green cloth"}, {"step": 7, "subtask": "grasp eggplant"}, {"step": 8, "subtask": "move grasped eggplant to the table"}, {"step": 9, "subtask": "release eggplant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2598fd4c6c2589d34bf8a1d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0222/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_39_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 222, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_39_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 379, "role": "error_localization"}}
{"state": {"task": "take the orange spoon to the right side of the burner", "plan": [{"step": 1, "subtask": "pick up orange spoon"}, {"step": 2, "subtask": "move grasped orange spoon to the right side of the burner"}, {"step": 3, "subtask": "release orange spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a4f5aa85aa9812d84d9c123f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0223/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_17_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 223, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_17_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 255, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tap handle"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp left burner"}, {"step": 10, "subtask": "move grasped object to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp left burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner knob"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0224/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 224, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 581, "role": "error_localization"}}
{"state": {"task": "Put spatula on plate, then on sink, then back on plate, then remove another object from drawer and place it on the sink.", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula to plate"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to sink"}, {"step": 6, "subtask": "release spatula"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to plate"}, {"step": 9, "subtask": "move it back to the original position"}, {"step": 10, "subtask": "release spatula"}, {"step": 11, "subtask": "grasp drawer handle"}, {"step": 12, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 13, "subtask": "release drawer handle"}, {"step": 14, "subtask": "grasp object"}, {"step": 15, "subtask": "move grasped object to sink"}, {"step": 16, "subtask": "release object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9f3ffe0b1b248aeee46759a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0225/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 225, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "2", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 492, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the lower side of the right stove, then move it to the upper side, then back to the lower side, and finally to the right side of the stove.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower side of the right stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper side of the right stove"}, {"step": 6, "subtask": "move it back to the original position"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to the lower side of the right stove"}, {"step": 10, "subtask": "release silver pot"}, {"step": 11, "subtask": "grasp silver pot"}, {"step": 12, "subtask": "move grasped silver pot to the right side of the stove"}, {"step": 13, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5fb632f98ec8ac70cfe3516.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0226/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_03_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 226, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_03_train", "episode_id": "17", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 472, "role": "error_localization"}}
{"state": {"task": "move the blue spoon to the top right corner of the stove, then to the top left corner, then to the bottom right corner, then back to the top right corner", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the top right corner of the stove"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the top left corner of the stove"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "release blue spoon"}, {"step": 9, "subtask": "move grasped blue spoon to the bottom right corner of the stove"}, {"step": 10, "subtask": "release blue spoon"}, {"step": 11, "subtask": "grasp blue spoon"}, {"step": 12, "subtask": "move grasped blue spoon to the top right corner of the stove"}, {"step": 13, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8dc55a1db643dd8960bc9dcc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0227/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_95_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 227, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_95_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 466, "role": "error_localization"}}
{"state": {"task": "Take the potato, move it to the right side of the stove, then to the left, then back to the right.", "plan": [{"step": 1, "subtask": "grasp potato"}, {"step": 2, "subtask": "move grasped potato to the right side of the stove"}, {"step": 3, "subtask": "release potato"}, {"step": 4, "subtask": "grasp stove"}, {"step": 5, "subtask": "move grasped potato to the left side of the stove"}, {"step": 6, "subtask": "release potato"}, {"step": 7, "subtask": "grasp stove"}, {"step": 8, "subtask": "move grasped potato to the right side of the stove"}, {"step": 9, "subtask": "release potato"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0228/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 228, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 369, "role": "error_localization"}}
{"state": {"task": "Remove the spoon from the orange pot, then place it on the table, pick it up again, and return it to the orange pot.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of orange pot"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to orange pot"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to table"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4b15d6af50f2480076658224.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0229/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 229, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "place the blue fork below the right burner, then to the left burner, then back below the right burner", "plan": [{"step": 1, "subtask": "grasp blue fork"}, {"step": 2, "subtask": "move grasped blue fork below the right burner"}, {"step": 3, "subtask": "release blue fork"}, {"step": 4, "subtask": "grasp blue fork"}, {"step": 5, "subtask": "move grasped blue fork to the left burner"}, {"step": 6, "subtask": "release blue fork"}, {"step": 7, "subtask": "grasp blue fork"}, {"step": 8, "subtask": "move grasped blue fork below the right burner"}, {"step": 9, "subtask": "release blue fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3bdf5023e0932785300f5e01.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0230/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_107_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 230, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_107_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "move the blue ladle to the left side of the burner, then to the right side, then to the front, then to the back", "plan": [{"step": 1, "subtask": "grasp blue ladle"}, {"step": 2, "subtask": "move grasped blue ladle to the left side of the burner"}, {"step": 3, "subtask": "release blue ladle"}, {"step": 4, "subtask": "grasp blue ladle"}, {"step": 5, "subtask": "move grasped blue ladle to the right side of the burner"}, {"step": 6, "subtask": "release blue ladle"}, {"step": 7, "subtask": "grasp blue ladle"}, {"step": 8, "subtask": "move grasped blue ladle to the front"}, {"step": 9, "subtask": "release blue ladle"}, {"step": 10, "subtask": "grasp blue ladle"}, {"step": 11, "subtask": "move grasped blue ladle to the back"}, {"step": 12, "subtask": "release blue ladle"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1537f3e7d28e2e69dbd822ba.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0231/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_27_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 231, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_27_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 444, "role": "error_localization"}}
{"state": {"task": "Put the yellow towel on the left burner, then move it to the right burner, then fold it neatly on the counter.", "plan": [{"step": 1, "subtask": "grasp yellow towel"}, {"step": 2, "subtask": "move grasped yellow towel to the right burner"}, {"step": 3, "subtask": "release yellow towel"}, {"step": 4, "subtask": "grasp yellow towel"}, {"step": 5, "subtask": "move grasped yellow towel to the left burner"}, {"step": 6, "subtask": "release yellow towel"}, {"step": 7, "subtask": "grasp yellow towel"}, {"step": 8, "subtask": "move grasped yellow towel to the counter"}, {"step": 9, "subtask": "fold grasped yellow towel"}, {"step": 10, "subtask": "release yellow towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70c5fe5722bd30d78ae38041.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0232/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 232, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "0", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 386, "role": "error_localization"}}
{"state": {"task": "move the cloth to top left, then fold it, then move it to bottom right, then unfold it", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to bottom left"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold grasped cloth"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to bottom right"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "unfold grasped cloth"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/10cbb8db026437d74d79a7b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0233/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 233, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 399, "role": "error_localization"}}
{"state": {"task": "put the cloth on the table to the burner", "plan": [{"step": 1, "subtask": "pick up cloth"}, {"step": 2, "subtask": "move grasped cloth to the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "pick up cloth"}, {"step": 5, "subtask": "move grasped cloth to the burner"}, {"step": 6, "subtask": "move it back to the original position"}, {"step": 7, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6289f95cd3b7f3e99705c663.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0234/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 234, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train", "episode_id": "5", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 313, "role": "error_localization"}}
{"state": {"task": "Move the blue scoop to the top left side of the stove, then to the top right side, then to the bottom left side, then to the bottom right side.", "plan": [{"step": 1, "subtask": "grasp blue scoop"}, {"step": 2, "subtask": "move grasped blue scoop to the top left side of the stove"}, {"step": 3, "subtask": "release blue scoop"}, {"step": 4, "subtask": "grasp blue scoop"}, {"step": 5, "subtask": "move grasped blue scoop to the top right side of the stove"}, {"step": 6, "subtask": "release blue scoop"}, {"step": 7, "subtask": "grasp blue scoop"}, {"step": 8, "subtask": "release blue scoop"}, {"step": 9, "subtask": "move grasped blue scoop to the bottom left side of the stove"}, {"step": 10, "subtask": "release blue scoop"}, {"step": 11, "subtask": "grasp blue scoop"}, {"step": 12, "subtask": "move grasped blue scoop to the bottom right side of the stove"}, {"step": 13, "subtask": "release blue scoop"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4330bced814cab052c9319be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0235/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 235, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "10", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 465, "role": "error_localization"}}
{"state": {"task": "pick up the banana and put it between the burners, then remove it and place it on the countertop, then pick it up again and return it between the burners", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana between the burners"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped banana to countertop"}, {"step": 6, "subtask": "release banana"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped banana between the burners"}, {"step": 9, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/125b5342fba4fb88cb682373.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0236/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 236, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 370, "role": "error_localization"}}
{"state": {"task": "Move the yellow knife to the right bottom of the burner, then to the top left, then back to the right bottom, and finally to the center.", "plan": [{"step": 1, "subtask": "grasp yellow knife"}, {"step": 2, "subtask": "move grasped yellow knife to the right bottom of the burner"}, {"step": 3, "subtask": "release yellow knife"}, {"step": 4, "subtask": "grasp yellow knife"}, {"step": 5, "subtask": "move grasped yellow knife to the top left"}, {"step": 6, "subtask": "release yellow knife"}, {"step": 7, "subtask": "grasp yellow knife"}, {"step": 8, "subtask": "move grasped yellow knife to the right bottom of the burner"}, {"step": 9, "subtask": "release yellow knife"}, {"step": 10, "subtask": "grasp yellow knife"}, {"step": 11, "subtask": "move grasped yellow knife to the center"}, {"step": 12, "subtask": "release yellow knife"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/de3a0c877178726a880868d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0237/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 237, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 435, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally back to the top right corner.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "move grasped cloth to top right corner"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "grasp cloth"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0238/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 238, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 424, "role": "error_localization"}}
{"state": {"task": "put the cans in the left cupboard, put the bread on the table, and put the red towel, and the orange bowl in the sink", "plan": [{"step": 1, "subtask": "grasp bottom cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to left cupboard"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp another can"}, {"step": 8, "subtask": "move grasped can to left cupboard"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped bread to table"}, {"step": 12, "subtask": "release bread"}, {"step": 13, "subtask": "grasp red towel"}, {"step": 14, "subtask": "move grasped red towel to sink"}, {"step": 15, "subtask": "release red towel"}, {"step": 16, "subtask": "grasp orange bowl"}, {"step": 17, "subtask": "move grasped orange bowl to sink"}, {"step": 18, "subtask": "release orange bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0239/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 239, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 523, "role": "error_localization"}}
{"state": {"task": "Move the red fork to the right side of the stove, then to the left side, then back to the right side.", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "grasp red fork"}, {"step": 5, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 6, "subtask": "release red fork"}, {"step": 7, "subtask": "grasp red fork"}, {"step": 8, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 9, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0240/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 240, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 378, "role": "error_localization"}}
{"state": {"task": "put the stuffed toy on the right bottom side of the table", "plan": [{"step": 1, "subtask": "pick up stuffed toy"}, {"step": 2, "subtask": "move grasped stuffed toy to the right bottom side of the table"}, {"step": 3, "subtask": "release stuffed toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a36e3a8487260877ca19c80.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0241/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train/36", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 241, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train", "episode_id": "36", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 257, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up bowl"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up corn"}, {"step": 5, "subtask": "release corn"}, {"step": 6, "subtask": "move grasped corn to sink"}, {"step": 7, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4564bea6b84939cd0266e47d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0242/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 242, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 302, "role": "error_localization"}}
{"state": {"task": "move the can to the center of the table, then to the front left, back right, and front right corners.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the center of the table"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the front left corner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the back right corner"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "move grasped can to the front right corner"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/577bf5296cef880c2fd84381.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0243/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 243, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 416, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp left burner knob"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp bowl"}, {"step": 10, "subtask": "move grasped bowl to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp left burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0244/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 244, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 580, "role": "error_localization"}}
{"state": {"task": "Remove blueberries from plate, place in bowl, then transfer to sink, then return to plate.", "plan": [{"step": 1, "subtask": "grasp blueberries"}, {"step": 2, "subtask": "move grasped blueberries to bowl"}, {"step": 3, "subtask": "release blueberries"}, {"step": 4, "subtask": "grasp blueberries"}, {"step": 5, "subtask": "move grasped blueberries to sink"}, {"step": 6, "subtask": "release blueberries"}, {"step": 7, "subtask": "grasp blueberries"}, {"step": 8, "subtask": "move grasped blueberries to plate"}, {"step": 9, "subtask": "release blueberries"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50178fd78a775935092228cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0245/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 245, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "move the red object to the bottom right of the stove, then to the top left of the stove, then back to the bottom right", "plan": [{"step": 1, "subtask": "grasp red object"}, {"step": 2, "subtask": "move grasped red object to the bottom right of the stove"}, {"step": 3, "subtask": "release red object"}, {"step": 4, "subtask": "grasp red object"}, {"step": 5, "subtask": "move grasped red object to the top left of the stove"}, {"step": 6, "subtask": "release red object"}, {"step": 7, "subtask": "grasp red object"}, {"step": 8, "subtask": "move grasped red object to the bottom right of the stove"}, {"step": 9, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9ca03d1ed51d690c9da373fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0246/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 246, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 381, "role": "error_localization"}}
{"state": {"task": "Move the silver pot on top of the green cloth, then place it on the blue mat, then return it to the cloth.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to blue mat"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dfd421c8d0c3cbba5e2cc99.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0247/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 247, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 373, "role": "error_localization"}}
{"state": {"task": "move the red bell pepper to the right side of burner, then to the front left, then to the back right, then back to the right side", "plan": [{"step": 1, "subtask": "grasp red bell pepper"}, {"step": 2, "subtask": "move grasped red bell pepper to the right side of burner"}, {"step": 3, "subtask": "release red bell pepper"}, {"step": 4, "subtask": "grasp red bell pepper"}, {"step": 5, "subtask": "move grasped red bell pepper to the front left"}, {"step": 6, "subtask": "release red bell pepper"}, {"step": 7, "subtask": "grasp red bell pepper"}, {"step": 8, "subtask": "move grasped red bell pepper to the back right"}, {"step": 9, "subtask": "release red bell pepper"}, {"step": 10, "subtask": "grasp red bell pepper"}, {"step": 11, "subtask": "move grasped red bell pepper to the right side of burner"}, {"step": 12, "subtask": "return it to its original position"}, {"step": 13, "subtask": "release red bell pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2f70c49451ef8db101e323c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0248/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 248, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train", "episode_id": "11", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 465, "role": "error_localization"}}
{"state": {"task": "Open the lip of the silver pot, move to the right upper side, then tilt it slightly to the left, then return it to its original position.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the right upper side"}, {"step": 3, "subtask": "rotate grasped silver pot slightly to the left"}, {"step": 4, "subtask": "move grasped silver pot to original position"}, {"step": 5, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/21e8524b6b602ef4ee0ea89b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0249/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 249, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 313, "role": "error_localization"}}
{"state": {"task": "move the croissant to the upper right side of the stove, then to the lower right side, then to the upper left side, then back to the upper right side", "plan": [{"step": 1, "subtask": "grasp croissant"}, {"step": 2, "subtask": "move grasped croissant to the upper right side of the stove"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release croissant"}, {"step": 5, "subtask": "grasp croissant"}, {"step": 6, "subtask": "move grasped croissant to the lower right side of the stove"}, {"step": 7, "subtask": "release croissant"}, {"step": 8, "subtask": "grasp croissant"}, {"step": 9, "subtask": "move grasped croissant to the upper left side of the stove"}, {"step": 10, "subtask": "release croissant"}, {"step": 11, "subtask": "grasp croissant"}, {"step": 12, "subtask": "move grasped croissant to the upper right side of the stove"}, {"step": 13, "subtask": "release croissant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2fa3823f2d16818c90494012.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0250/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 250, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train", "episode_id": "14", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 470, "role": "error_localization"}}
{"state": {"task": "place the box in the pot on the left burner", "plan": [{"step": 1, "subtask": "pick up box"}, {"step": 2, "subtask": "move grasped box to pot"}, {"step": 3, "subtask": "release box"}, {"step": 4, "subtask": "pick up pot"}, {"step": 5, "subtask": "move grasped pot to left burner"}, {"step": 6, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dd7e6cb40aec37575c113f26.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0251/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 251, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 294, "role": "error_localization"}}
{"state": {"task": "put the napkin on the bottom right of the stove, then move it to the bottom left, then to the top right, then to the top left", "plan": [{"step": 1, "subtask": "grasp napkin"}, {"step": 2, "subtask": "move grasped napkin to the bottom right of the stove"}, {"step": 3, "subtask": "release napkin"}, {"step": 4, "subtask": "grasp napkin"}, {"step": 5, "subtask": "move grasped napkin to the bottom left of the stove"}, {"step": 6, "subtask": "release napkin"}, {"step": 7, "subtask": "grasp napkin"}, {"step": 8, "subtask": "move grasped napkin to the top right of the stove"}, {"step": 9, "subtask": "release napkin"}, {"step": 10, "subtask": "grasp napkin"}, {"step": 11, "subtask": "move grasped napkin to the top left of the stove"}, {"step": 12, "subtask": "release napkin"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d426044ecd9e4dfa01e60abf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0252/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_05_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 252, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_05_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 443, "role": "error_localization"}}
{"state": {"task": "move the cloth to the right of the burner, then fold it, then place it to the left of the burner, then return it to the right of the burner", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the right of the burner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "rotate grasped cloth"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to the left of the burner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth to the right of the burner"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8f7ffe5092813e028dc9f080.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0253/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 253, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 423, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from top right to bottom left", "plan": [{"step": 1, "subtask": "move grasped cloth from top right to bottom left"}, {"step": 2, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9441a43315c831de0b097470.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0254/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_06_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 254, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_06_train", "episode_id": "14", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 234, "role": "error_localization"}}
{"state": {"task": "Move the yellow napkin to the upper right of the table, then to the lower left, then back to the upper right, and finally to the center.", "plan": [{"step": 1, "subtask": "grasp yellow napkin"}, {"step": 2, "subtask": "move grasped yellow napkin to the upper right of the table"}, {"step": 3, "subtask": "release yellow napkin"}, {"step": 4, "subtask": "grasp yellow napkin"}, {"step": 5, "subtask": "move grasped yellow napkin to the lower left of the table"}, {"step": 6, "subtask": "release yellow napkin"}, {"step": 7, "subtask": "grasp yellow napkin"}, {"step": 8, "subtask": "move grasped yellow napkin to the upper right of the table"}, {"step": 9, "subtask": "release yellow napkin"}, {"step": 10, "subtask": "grasp yellow napkin"}, {"step": 11, "subtask": "move grasped yellow napkin to the center of the table"}, {"step": 12, "subtask": "release yellow napkin"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1fdee2219c6a4f0291bcaff7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0255/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_25_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 255, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_25_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "Remove the silver lid from the silver pot, then place it on the table, and put the silver pot back under the sink.", "plan": [{"step": 1, "subtask": "grasp silver lid"}, {"step": 2, "subtask": "move grasped silver lid out of silver pot"}, {"step": 3, "subtask": "move grasped silver pot to under sink"}, {"step": 4, "subtask": "grasp silver lid"}, {"step": 5, "subtask": "move grasped silver lid to table"}, {"step": 6, "subtask": "release silver lid"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "release silver lid"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/be32103799c792ede7b439c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0256/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 256, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "0", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 367, "role": "error_localization"}}
{"state": {"task": "put the silver pot lid on the top of the silver pot, then remove it, then place it on the side of the silver pot", "plan": [{"step": 1, "subtask": "grasp silver pot lid"}, {"step": 2, "subtask": "release silver pot lid"}, {"step": 3, "subtask": "move grasped silver pot lid to the side of the silver pot"}, {"step": 4, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/58e7d8debd78055fa8271394.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0257/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 257, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "14", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 290, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl, then take bowl out, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "release bowl"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to cabinet"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp another spoon"}, {"step": 11, "subtask": "move grasped spoon to bowl"}, {"step": 12, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0258/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 258, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 404, "role": "error_localization"}}
{"state": {"task": "Move the red spoon to the bottom left from the top right of the burner, then to the top left, then to the bottom right, then back to the top right of the burner.", "plan": [{"step": 1, "subtask": "grasp red spoon"}, {"step": 2, "subtask": "move grasped red spoon to the bottom left of the burner"}, {"step": 3, "subtask": "release red spoon"}, {"step": 4, "subtask": "grasp red spoon"}, {"step": 5, "subtask": "move grasped red spoon to the top left of the burner"}, {"step": 6, "subtask": "release red spoon"}, {"step": 7, "subtask": "grasp red spoon"}, {"step": 8, "subtask": "move grasped red spoon to the bottom right of the burner"}, {"step": 9, "subtask": "release red spoon"}, {"step": 10, "subtask": "grasp red spoon"}, {"step": 11, "subtask": "move grasped red spoon to the top right of the burner"}, {"step": 12, "subtask": "release red spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b36c36489b429f93624d2a6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0259/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_00_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 259, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_00_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 449, "role": "error_localization"}}
{"state": {"task": "Put the blue spoon on the top of the green towel, then move it to the left edge, then to the right edge, then back to the center.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the top of the green towel"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the left edge of the green towel"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to the right edge of the green towel"}, {"step": 9, "subtask": "release blue spoon"}, {"step": 10, "subtask": "grasp blue spoon"}, {"step": 11, "subtask": "move grasped blue spoon to the center of the green towel"}, {"step": 12, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c728ac183e0042e2196d76b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0260/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_31_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 260, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_31_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "put the banana on the napkin on the left side of the table", "plan": [{"step": 1, "subtask": "pick up banana"}, {"step": 2, "subtask": "move grasped banana to the left side of the table"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "pick up napkin"}, {"step": 5, "subtask": "move grasped napkin to the left side of the table"}, {"step": 6, "subtask": "release napkin"}, {"step": 7, "subtask": "pick up banana"}, {"step": 8, "subtask": "move grasped banana to napkin"}, {"step": 9, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91d129be15c2df2fd7b2cbe9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0261/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 261, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "remove the wooden spatula to the basket and place it from the side of the right burner", "plan": [{"step": 1, "subtask": "pick up wooden spatula"}, {"step": 2, "subtask": "move grasped wooden spatula to the basket"}, {"step": 3, "subtask": "release wooden spatula"}, {"step": 4, "subtask": "move grasped wooden spatula from the side of the right burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b369b2e6e20494b0b44c6704.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0262/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_03_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 262, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_03_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 286, "role": "error_localization"}}
{"state": {"task": "remove the wooden spatula to the basket and place it from the side of the right burner", "plan": [{"step": 1, "subtask": "pick up wooden spatula"}, {"step": 2, "subtask": "release wooden spatula"}, {"step": 3, "subtask": "move grasped wooden spatula to the basket"}, {"step": 4, "subtask": "release wooden spatula"}, {"step": 5, "subtask": "move grasped wooden spatula from the side of the right burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b369b2e6e20494b0b44c6704.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0263/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_03_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 263, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_03_train", "episode_id": "0", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 302, "role": "error_localization"}}
{"state": {"task": "take banana out of colander", "plan": [{"step": 1, "subtask": "pick up colander"}, {"step": 2, "subtask": "move grasped colander to a stable surface"}, {"step": 3, "subtask": "release colander"}, {"step": 4, "subtask": "pick up banana"}, {"step": 5, "subtask": "move grasped banana away from colander"}, {"step": 6, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/71968079ee95315f5cd44abe.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0264/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 264, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 296, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put the yellow can to the upper right corner", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp tomato can"}, {"step": 11, "subtask": "move grasped can to upper right corner"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0265/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 265, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 428, "role": "error_localization"}}
{"state": {"task": "remove cucumber from cup, then place the cucumber on the table, then return it to the cup", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber out of cup"}, {"step": 3, "subtask": "release cucumber"}, {"step": 4, "subtask": "grasp cucumber"}, {"step": 5, "subtask": "move grasped cucumber to table"}, {"step": 6, "subtask": "release cucumber"}, {"step": 7, "subtask": "grasp cucumber"}, {"step": 8, "subtask": "move grasped cucumber to cup"}, {"step": 9, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23b9766f583bf6909b474dc1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0266/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/34", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 266, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "34", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 350, "role": "error_localization"}}
{"state": {"task": "move the green and yellow corn to the lower right corner of the stove", "plan": [{"step": 1, "subtask": "pick up green corn"}, {"step": 2, "subtask": "move grasped green corn to the lower right corner of the stove"}, {"step": 3, "subtask": "release green corn"}, {"step": 4, "subtask": "pick up yellow corn"}, {"step": 5, "subtask": "move grasped yellow corn to the lower right corner of the stove"}, {"step": 6, "subtask": "release yellow corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/54526650fc97df9303b3a813.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0267/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 267, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_10_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 315, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the upper middle of the table, then to the lower left corner, then back to the center", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper middle of the table"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "grasp silver pot"}, {"step": 7, "subtask": "move grasped silver pot to the center of the table"}, {"step": 8, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8d6dfedb78b3811df610b491.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0268/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_12_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 268, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_12_train", "episode_id": "14", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "Move the yellow object to the bottom right edge of burners, then to the top left edge, then back to the bottom right edge.", "plan": [{"step": 1, "subtask": "grasp yellow object"}, {"step": 2, "subtask": "move grasped yellow object to bottom right edge of burners"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release yellow object"}, {"step": 5, "subtask": "grasp yellow object"}, {"step": 6, "subtask": "move grasped yellow object to top left edge of burners"}, {"step": 7, "subtask": "release yellow object"}, {"step": 8, "subtask": "grasp yellow object"}, {"step": 9, "subtask": "move grasped yellow object to bottom right edge of burners"}, {"step": 10, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5256d032f07b2e7b5bb8362c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0269/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 269, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train", "episode_id": "19", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 401, "role": "error_localization"}}
{"state": {"task": "Remove the yellow scrubber from the green cloth, then place the scrubber on the table, then replace it back on the cloth.", "plan": [{"step": 1, "subtask": "grasp yellow scrubber"}, {"step": 2, "subtask": "move grasped yellow scrubber out of green cloth"}, {"step": 3, "subtask": "release yellow scrubber"}, {"step": 4, "subtask": "grasp yellow scrubber"}, {"step": 5, "subtask": "move grasped yellow scrubber to green cloth"}, {"step": 6, "subtask": "release yellow scrubber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c99f5cb2750d538b55f158b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0270/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 270, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 324, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up bowl"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up corn"}, {"step": 5, "subtask": "move grasped corn to sink"}, {"step": 6, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4564bea6b84939cd0266e47d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0271/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 271, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 288, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from top of the green cloth, place it on the table, then return it to the cloth, finally lift the cloth to reveal what's underneath.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to green cloth"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp green cloth"}, {"step": 8, "subtask": "move grasped green cloth up"}, {"step": 9, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/63fcd9930f4272aa1004534d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0272/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 272, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "move the blue spatula to the left side of the burner, then to the right side, then to the center, then back to the left side", "plan": [{"step": 1, "subtask": "grasp blue spatula"}, {"step": 2, "subtask": "move grasped blue spatula to the left side of the burner"}, {"step": 3, "subtask": "release blue spatula"}, {"step": 4, "subtask": "grasp blue spatula"}, {"step": 5, "subtask": "move grasped blue spatula to the right side of the burner"}, {"step": 6, "subtask": "release blue spatula"}, {"step": 7, "subtask": "grasp blue spatula"}, {"step": 8, "subtask": "move grasped blue spatula to the right side of the burner"}, {"step": 9, "subtask": "release blue spatula"}, {"step": 10, "subtask": "grasp blue spatula"}, {"step": 11, "subtask": "move grasped blue spatula to the left side of the burner"}, {"step": 12, "subtask": "release blue spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14f4a698e896767d3f97c4b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0273/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 273, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "put the silver pot to the center of the burner", "plan": [{"step": 1, "subtask": "pick up silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the center of the burner"}, {"step": 3, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a11e82fa3b7b4b4505aa0993.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0274/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 274, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 253, "role": "error_localization"}}
{"state": {"task": "remove the corn from the silver pot", "plan": [{"step": 1, "subtask": "pick up silver pot"}, {"step": 2, "subtask": "move grasped silver pot upright"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "pick up corn"}, {"step": 5, "subtask": "move grasped corn away from silver pot"}, {"step": 6, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8ea5130969d36de31f2e4cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0275/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_15_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 275, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_15_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 294, "role": "error_localization"}}
{"state": {"task": "put the brown lid on the brown pot, then remove the lid and place it on the table, then put the lid back on the pot, then lift the pot and lid together", "plan": [{"step": 1, "subtask": "grasp brown lid"}, {"step": 2, "subtask": "move grasped brown lid to brown pot"}, {"step": 3, "subtask": "release brown lid"}, {"step": 4, "subtask": "grasp brown lid"}, {"step": 5, "subtask": "move grasped brown lid out"}, {"step": 6, "subtask": "move grasped brown lid to table"}, {"step": 7, "subtask": "release brown lid"}, {"step": 8, "subtask": "grasp brown lid"}, {"step": 9, "subtask": "move grasped brown lid to brown pot"}, {"step": 10, "subtask": "release brown lid"}, {"step": 11, "subtask": "grasp brown pot"}, {"step": 12, "subtask": "move grasped brown pot up"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3e15ddef653eba328a364d00.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0276/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 276, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_22_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "Move the red cloth to the lower left corner of the stove, then to the upper left corner, then to the upper right corner, then back to the lower left corner.", "plan": [{"step": 1, "subtask": "grasp red cloth"}, {"step": 2, "subtask": "move grasped red cloth to the upper left corner of the stove"}, {"step": 3, "subtask": "release red cloth"}, {"step": 4, "subtask": "grasp red cloth"}, {"step": 5, "subtask": "move grasped red cloth to the lower left corner of the stove"}, {"step": 6, "subtask": "release red cloth"}, {"step": 7, "subtask": "grasp red cloth"}, {"step": 8, "subtask": "move grasped red cloth to the upper right corner of the stove"}, {"step": 9, "subtask": "release red cloth"}, {"step": 10, "subtask": "grasp red cloth"}, {"step": 11, "subtask": "move grasped red cloth to the lower left corner of the stove"}, {"step": 12, "subtask": "release red cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0277/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 277, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 450, "role": "error_localization"}}
{"state": {"task": "move the cloth from left to right, then fold it in half, then unfold it, then move it from right to left", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth from left to right"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold grasped cloth in half"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "unfold grasped cloth"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth from right to left"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0e2b1d9947457624b7752da5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0278/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_03_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 278, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_03_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 407, "role": "error_localization"}}
{"state": {"task": "Move the orange toy to the left side of the table, then to the center, then back to the left side.", "plan": [{"step": 1, "subtask": "grasp orange toy"}, {"step": 2, "subtask": "grasp orange toy"}, {"step": 3, "subtask": "release orange toy"}, {"step": 4, "subtask": "move grasped orange toy to the left side of the table"}, {"step": 5, "subtask": "move grasped orange toy to the center of the table"}, {"step": 6, "subtask": "release orange toy"}, {"step": 7, "subtask": "grasp orange toy"}, {"step": 8, "subtask": "move grasped orange toy to the left side of the table"}, {"step": 9, "subtask": "release orange toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0279/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 279, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 376, "role": "error_localization"}}
{"state": {"task": "remove the silver bowl from the left burner, then place it on the counter, then move it to the right burner, then return it to the left burner", "plan": [{"step": 1, "subtask": "grasp silver bowl"}, {"step": 2, "subtask": "move grasped silver bowl out of left burner"}, {"step": 3, "subtask": "release silver bowl"}, {"step": 4, "subtask": "grasp silver bowl"}, {"step": 5, "subtask": "move grasped silver bowl to right burner"}, {"step": 6, "subtask": "release silver bowl"}, {"step": 7, "subtask": "grasp silver bowl"}, {"step": 8, "subtask": "move grasped silver bowl to left burner"}, {"step": 9, "subtask": "release silver bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8bfdb2434a17a7dd6ebe4bb8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0280/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 280, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 374, "role": "error_localization"}}
{"state": {"task": "put strawberry in bowl, then take bowl out of sink onto the table, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped object to bowl"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "move grasped bowl to the table"}, {"step": 7, "subtask": "release bowl"}, {"step": 8, "subtask": "grasp cabinet door handle"}, {"step": 9, "subtask": "move grasped door handle to open position"}, {"step": 10, "subtask": "release door handle"}, {"step": 11, "subtask": "grasp spoon"}, {"step": 12, "subtask": "move grasped spoon to cabinet"}, {"step": 13, "subtask": "release spoon"}, {"step": 14, "subtask": "grasp fork"}, {"step": 15, "subtask": "move grasped object to bowl"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0281/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 281, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 481, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the right top burner, then to the left top burner, then back to the right top burner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the right top burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the left top burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the right top burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8fd7aa3066f81bb9a4d4371d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0282/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 282, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "Move the silver pot on the top of the orange towel, then place it next to the silver fork, then back on the orange towel.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the top of the orange towel"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top of the orange towel"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the top of the silver fork"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bb027c35653a12daa376ee60.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0283/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 283, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 381, "role": "error_localization"}}
{"state": {"task": "move the can to the center of the table, then to the front left, back right, and front right corners.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the center of the table"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the back left corner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the back right corner"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "move grasped can to the front right corner"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/577bf5296cef880c2fd84381.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0284/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 284, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 416, "role": "error_localization"}}
{"state": {"task": "Put the cup to the counter from the sink, then to the stove, then to the fridge handle, then back to the counter.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to fridge handle"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to counter"}, {"step": 12, "subtask": "move it back to the original position"}, {"step": 13, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b321bf8039405b2c9c4ac518.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0285/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/67", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 285, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "67", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 427, "role": "error_localization"}}
{"state": {"task": "Move the silver lid to the top of the silver pot, then remove the red cup from the drawer and place it next to the pot, then move the lid to the cup.", "plan": [{"step": 1, "subtask": "grasp silver lid"}, {"step": 2, "subtask": "move grasped silver lid to the top of the silver pot"}, {"step": 3, "subtask": "release silver lid"}, {"step": 4, "subtask": "grasp drawer handle"}, {"step": 5, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 6, "subtask": "release drawer handle"}, {"step": 7, "subtask": "grasp red cup"}, {"step": 8, "subtask": "move grasped red cup next to the pot"}, {"step": 9, "subtask": "release red cup"}, {"step": 10, "subtask": "grasp silver lid"}, {"step": 11, "subtask": "move grasped silver lid to the red cup"}, {"step": 12, "subtask": "release silver lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3cb0f195916bba5d74618c3f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0286/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 286, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 439, "role": "error_localization"}}
{"state": {"task": "Place the blue spoon on top of the orange cloth, then move the spoon to the center of the table, then back to the cloth.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to center of the table"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to top of the orange cloth"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to center of the table"}, {"step": 9, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/90bf73d2486a48559acb5f8d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0287/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_67_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 287, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_67_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 376, "role": "error_localization"}}
{"state": {"task": "move the blue plastic cup to the top of the left burner, then to the top of the right burner, then back to the left burner, then next to the red cup on the table", "plan": [{"step": 1, "subtask": "grasp blue plastic cup"}, {"step": 2, "subtask": "move grasped blue plastic cup to the top of the left burner"}, {"step": 3, "subtask": "release blue plastic cup"}, {"step": 4, "subtask": "grasp blue plastic cup"}, {"step": 5, "subtask": "move grasped blue plastic cup to the top of the right burner"}, {"step": 6, "subtask": "release blue plastic cup"}, {"step": 7, "subtask": "grasp blue plastic cup"}, {"step": 8, "subtask": "move grasped blue plastic cup to the top of the left burner"}, {"step": 9, "subtask": "release blue plastic cup"}, {"step": 10, "subtask": "grasp blue plastic cup"}, {"step": 11, "subtask": "move grasped blue plastic cup next to the red cup on the table"}, {"step": 12, "subtask": "release blue plastic cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0cf2f19d527950a74478b2b0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0288/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_56_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 288, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_56_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 463, "role": "error_localization"}}
{"state": {"task": "Pick up the yellow object, place it below the right burner, then place it to the left burner, then return it to the right burner.", "plan": [{"step": 1, "subtask": "grasp yellow object"}, {"step": 2, "subtask": "move grasped yellow object below the right burner"}, {"step": 3, "subtask": "release yellow object"}, {"step": 4, "subtask": "grasp yellow object"}, {"step": 5, "subtask": "move grasped yellow object to the left burner"}, {"step": 6, "subtask": "release yellow object"}, {"step": 7, "subtask": "grasp yellow object"}, {"step": 8, "subtask": "move grasped yellow object to the right burner"}, {"step": 9, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac62285427547839b7682924.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0289/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 289, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 373, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up bowl"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up corn"}, {"step": 5, "subtask": "move grasped corn to sink"}, {"step": 6, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1fca3cb2b20369cb9958fcd1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0290/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 290, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 288, "role": "error_localization"}}
{"state": {"task": "remove cup from plate, place it on the table, then move the plate to the sink, then return the cup to the plate", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to table"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp plate"}, {"step": 8, "subtask": "move grasped plate to sink"}, {"step": 9, "subtask": "release plate"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to plate"}, {"step": 12, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6b9f1ba9d9e66b13306a4f11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0291/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/27", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 291, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "27", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 407, "role": "error_localization"}}
{"state": {"task": "Move one of the cans to the upper left corner of the left burner, then to the upper right corner, then to the lower left corner, then back to the center.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the upper left corner of the left burner"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the upper right corner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the lower left corner"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "move grasped can to the center"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c850847225497b2c7a0b7727.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0292/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 292, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 428, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom right corner of the burner, then to the top left, then to the center", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom right corner of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top left corner of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center of the burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c90750c2ccb5c44fcd5f9de5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0293/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 293, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 378, "role": "error_localization"}}
{"state": {"task": "take the yellow bottle with black label and put it on the right burner, then to the left burner, then back to the right burner", "plan": [{"step": 1, "subtask": "grasp yellow bottle with black label"}, {"step": 2, "subtask": "move grasped yellow bottle with black label to the left burner"}, {"step": 3, "subtask": "release yellow bottle with black label"}, {"step": 4, "subtask": "grasp yellow bottle with black label"}, {"step": 5, "subtask": "move grasped yellow bottle with black label to the left burner"}, {"step": 6, "subtask": "release yellow bottle with black label"}, {"step": 7, "subtask": "grasp yellow bottle with black label"}, {"step": 8, "subtask": "move grasped yellow bottle with black label to the left burner"}, {"step": 9, "subtask": "release yellow bottle with black label"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cf8e8869f198215689760a2f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0294/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 294, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 399, "role": "error_localization"}}
{"state": {"task": "move the red spoon to the right side of the stove, then to the left side, then back to the right side.", "plan": [{"step": 1, "subtask": "grasp red spoon"}, {"step": 2, "subtask": "move grasped red spoon to the right side of the stove"}, {"step": 3, "subtask": "release red spoon"}, {"step": 4, "subtask": "grasp red spoon"}, {"step": 5, "subtask": "move grasped red spoon to the left side of the stove"}, {"step": 6, "subtask": "release red spoon"}, {"step": 7, "subtask": "grasp red spoon"}, {"step": 8, "subtask": "move grasped red spoon to the right side of the stove"}, {"step": 9, "subtask": "release red spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bee6ca8c03cbe47ce25d1801.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0295/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_05_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 295, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_05_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 378, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the bottom center of the stove between the orange object and green cloth, then to the top left, then back to the center.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom center of the stove between the orange object and green cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top left"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center of the stove"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9c8d2f9423459fe8806acee1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0296/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_45_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 296, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_45_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "Move the yellow cloth from the middle front of the burners to behind the right side of the burners, then to the left side, then back to the middle front.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to behind the left side of the burners"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth to the left side of the burners"}, {"step": 6, "subtask": "release yellow cloth"}, {"step": 7, "subtask": "grasp yellow cloth"}, {"step": 8, "subtask": "move grasped yellow cloth to the middle front of the burners"}, {"step": 9, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b13a7904f6d38c43dac100f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0297/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 297, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 392, "role": "error_localization"}}
{"state": {"task": "Move the green spatula from the green cloth to center of the stove, then to the front left burner, then to the front right burner, then back to the center.", "plan": [{"step": 1, "subtask": "grasp green spatula"}, {"step": 2, "subtask": "move grasped green spatula to center of the stove"}, {"step": 3, "subtask": "release green spatula"}, {"step": 4, "subtask": "grasp green spatula"}, {"step": 5, "subtask": "move grasped green spatula to front left burner"}, {"step": 6, "subtask": "release green spatula"}, {"step": 7, "subtask": "grasp green spatula"}, {"step": 8, "subtask": "move grasped green spatula to front right burner"}, {"step": 9, "subtask": "release green spatula"}, {"step": 10, "subtask": "grasp green spatula"}, {"step": 11, "subtask": "move grasped green spatula to center of the stove"}, {"step": 12, "subtask": "release green spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/26fc795ac507115bdda01405.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0298/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 298, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 448, "role": "error_localization"}}
{"state": {"task": "Put the blue spoon on the top of the green towel, then move it to the left edge, then to the right edge, then back to the center.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the top of the green towel"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the left edge of the green towel"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to the left edge of the green towel"}, {"step": 9, "subtask": "release blue spoon"}, {"step": 10, "subtask": "grasp blue spoon"}, {"step": 11, "subtask": "move grasped blue spoon to the center of the green towel"}, {"step": 12, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c728ac183e0042e2196d76b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0299/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_31_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 299, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_31_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "Remove the blue cloth from the top edge of the cloth, then place it on the bottom edge, then return it to the top edge, then set it aside.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth away from the top edge of the cloth"}, {"step": 3, "subtask": "release blue cloth"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "move grasped blue cloth to the bottom edge of the cloth"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp blue cloth"}, {"step": 8, "subtask": "move grasped blue cloth to the top edge of the cloth"}, {"step": 9, "subtask": "release blue cloth"}, {"step": 10, "subtask": "grasp blue cloth"}, {"step": 11, "subtask": "move grasped blue cloth away from the top edge of the cloth"}, {"step": 12, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86116ee6494d87d8d320afb2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0300/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 300, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 446, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl, then take bowl out of sink onto the table, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "move grasped bowl to the table"}, {"step": 7, "subtask": "release bowl"}, {"step": 8, "subtask": "grasp oven door handle"}, {"step": 9, "subtask": "move grasped door handle to open position"}, {"step": 10, "subtask": "release door handle"}, {"step": 11, "subtask": "grasp spoon"}, {"step": 12, "subtask": "move grasped spoon to cabinet"}, {"step": 13, "subtask": "release spoon"}, {"step": 14, "subtask": "grasp fork"}, {"step": 15, "subtask": "move grasped object to bowl"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0301/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 301, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 482, "role": "error_localization"}}
{"state": {"task": "move the green object into the silver pot, then remove a lid from a nearby drawer and place it on top of the pot, then take the green object out of the pot", "plan": [{"step": 1, "subtask": "move grasped lid to top of silver pot"}, {"step": 2, "subtask": "move grasped green object to silver pot"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp drawer handle"}, {"step": 5, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 6, "subtask": "release drawer handle"}, {"step": 7, "subtask": "grasp lid"}, {"step": 8, "subtask": "grasp green object"}, {"step": 9, "subtask": "release lid"}, {"step": 10, "subtask": "grasp green object"}, {"step": 11, "subtask": "move grasped green object out of silver pot"}, {"step": 12, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0303/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 303, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 433, "role": "error_localization"}}
{"state": {"task": "Remove the blue cloth from the top edge of the cloth, then place it on the table, then return it to the cloth's bottom edge, then move it back to the top edge.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth out"}, {"step": 3, "subtask": "release blue cloth"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "move grasped blue cloth to table"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp blue cloth"}, {"step": 8, "subtask": "move grasped blue cloth to bottom edge of cloth"}, {"step": 9, "subtask": "release blue cloth"}, {"step": 10, "subtask": "grasp blue cloth"}, {"step": 11, "subtask": "move grasped blue cloth to top edge of cloth"}, {"step": 12, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dc4228d0728db133a5314b33.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0304/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/31", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 304, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "31", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 434, "role": "error_localization"}}
{"state": {"task": "remove the green spatula from the right burner, then place it on the left burner, then return it to the right burner", "plan": [{"step": 1, "subtask": "grasp green spatula"}, {"step": 2, "subtask": "move grasped green spatula out of right burner"}, {"step": 3, "subtask": "release green spatula"}, {"step": 4, "subtask": "grasp green spatula"}, {"step": 5, "subtask": "move grasped green spatula to left burner"}, {"step": 6, "subtask": "release green spatula"}, {"step": 7, "subtask": "grasp green spatula"}, {"step": 8, "subtask": "move grasped green spatula to right burner"}, {"step": 9, "subtask": "release green spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13acb27e9a90f1e76fe5bdf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0305/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_51_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 305, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_51_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 377, "role": "error_localization"}}
{"state": {"task": "move the cloth from the bottom right to the top left of the table, then fold it in half, then move it back to the bottom right", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the top left of the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold grasped cloth in half"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to the bottom right of the table"}, {"step": 9, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf568029e1d1a3734b86d995.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0306/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 306, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 369, "role": "error_localization"}}
{"state": {"task": "try to pull up the green cloth, then fold it, then put it inside the box", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth out"}, {"step": 3, "subtask": "rotate grasped green cloth"}, {"step": 4, "subtask": "move grasped green cloth to floor"}, {"step": 5, "subtask": "release green cloth"}, {"step": 6, "subtask": "grasp green cloth"}, {"step": 7, "subtask": "move grasped green cloth to outside box"}, {"step": 8, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fabe07a6742a375eeff13925.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0307/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 307, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 343, "role": "error_localization"}}
{"state": {"task": "close the drawer after putting the items inside it back in their proper places", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp object"}, {"step": 5, "subtask": "move grasped object to target"}, {"step": 6, "subtask": "release object"}, {"step": 7, "subtask": "grasp drawer handle"}, {"step": 8, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 9, "subtask": "release drawer handle"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d54b05fdb7cf5b7311fd3735.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0308/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 308, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 356, "role": "error_localization"}}
{"state": {"task": "close the bottle top", "plan": [{"step": 1, "subtask": "pick up bottle"}, {"step": 2, "subtask": "release bottle"}, {"step": 3, "subtask": "move grasped bottle to upright position"}, {"step": 4, "subtask": "release bottle"}, {"step": 5, "subtask": "pick up bottle top"}, {"step": 6, "subtask": "move grasped bottle top to the bottle's closed position"}, {"step": 7, "subtask": "release bottle top"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f06814aa0250a1830e2561dd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0309/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_20_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 309, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_20_train", "episode_id": "8", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 309, "role": "error_localization"}}
{"state": {"task": "Put the green toy on top of the purple cloth, then move it to the center, then to the edge, then back to the center.", "plan": [{"step": 1, "subtask": "grasp green toy"}, {"step": 2, "subtask": "move grasped green toy to top of purple cloth"}, {"step": 3, "subtask": "release green toy"}, {"step": 4, "subtask": "grasp green toy"}, {"step": 5, "subtask": "move grasped green toy to center of purple cloth"}, {"step": 6, "subtask": "release green toy"}, {"step": 7, "subtask": "grasp green toy"}, {"step": 8, "subtask": "move grasped green toy to edge of purple cloth"}, {"step": 9, "subtask": "release green toy"}, {"step": 10, "subtask": "grasp green toy"}, {"step": 11, "subtask": "move grasped green toy to center of purple cloth"}, {"step": 12, "subtask": "release green toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4fabef0ffb84d546246736c7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0310/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 310, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "Put the orange object above the right burner, then move it to the left burner, then back to the right burner.", "plan": [{"step": 1, "subtask": "grasp orange object"}, {"step": 2, "subtask": "move grasped orange object above the right burner"}, {"step": 3, "subtask": "release orange object"}, {"step": 4, "subtask": "grasp orange object"}, {"step": 5, "subtask": "move grasped orange object to the left burner"}, {"step": 6, "subtask": "release orange object"}, {"step": 7, "subtask": "grasp orange object"}, {"step": 8, "subtask": "move grasped orange object above the right burner"}, {"step": 9, "subtask": "release orange object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c746d0c4d469ec9dac6b3fcd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0311/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_19_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 311, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_19_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 368, "role": "error_localization"}}
{"state": {"task": "put cup from counter to sink", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to sink"}, {"step": 3, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6a8fe056cc5a0cd4a151e15.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0312/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/57", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 312, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "57", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 242, "role": "error_localization"}}
{"state": {"task": "pick up blue pen and put into drawer", "plan": [{"step": 1, "subtask": "pick up drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "pick up blue pen"}, {"step": 5, "subtask": "move grasped blue pen to drawer"}, {"step": 6, "subtask": "release blue pen"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4233b866de77b685595cea8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0313/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 313, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_blue_pen_and_put_into_drawer_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 300, "role": "error_localization"}}
{"state": {"task": "move the silver pot on top of the green cloth, then move it to the wooden tray, then back on top of the green cloth", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "release silver pot"}, {"step": 3, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to wooden tray"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dfd421c8d0c3cbba5e2cc99.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0314/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 314, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "3", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 375, "role": "error_localization"}}
{"state": {"task": "Move the red object to the left burner, then to the right burner, then back to the left burner.", "plan": [{"step": 1, "subtask": "grasp red object"}, {"step": 2, "subtask": "move grasped red object to left burner"}, {"step": 3, "subtask": "release red object"}, {"step": 4, "subtask": "grasp red object"}, {"step": 5, "subtask": "move grasped red object to right burner"}, {"step": 6, "subtask": "release red object"}, {"step": 7, "subtask": "grasp red object"}, {"step": 8, "subtask": "move grasped red object to left burner"}, {"step": 9, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9d8611db6f5ba9488cf9e6b4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0315/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 315, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 363, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from the orange cloth, place it on the table, then return it to the orange cloth, then move it to the green towel.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out of orange cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to orange cloth"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp orange cloth"}, {"step": 11, "subtask": "move grasped silver pot to the green towel"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0316/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 316, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 425, "role": "error_localization"}}
{"state": {"task": "remove the pink toy from the top of the drawer", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up pink toy"}, {"step": 3, "subtask": "release pink toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/93541c39e97072b2e745847e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0317/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 317, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 244, "role": "error_localization"}}
{"state": {"task": "put cup to counter from sink, then to the stove, then back to the counter, then to the table", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to counter"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "move grasped cup to table"}, {"step": 11, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aba7dfe4d7f485e002ab0ec2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0318/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 318, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 386, "role": "error_localization"}}
{"state": {"task": "move the blue spoon to the top right corner of the stove, then to the top left corner, then to the bottom right corner, then back to the top right corner", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the top right corner of the stove"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the top left corner of the stove"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to the bottom right corner of the stove"}, {"step": 9, "subtask": "release blue spoon"}, {"step": 10, "subtask": "grasp blue spoon"}, {"step": 11, "subtask": "move grasped blue spoon to the top right corner of the stove"}, {"step": 12, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8dc55a1db643dd8960bc9dcc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0319/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_95_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 319, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_95_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 450, "role": "error_localization"}}
{"state": {"task": "move the red spoon to the front left corner of the stove, then to the front right corner, then to the back left corner, then to the back right corner", "plan": [{"step": 1, "subtask": "grasp red spoon"}, {"step": 2, "subtask": "move grasped red spoon to the front left corner of the stove"}, {"step": 3, "subtask": "release red spoon"}, {"step": 4, "subtask": "grasp red spoon"}, {"step": 5, "subtask": "move grasped red spoon to the front right corner of the stove"}, {"step": 6, "subtask": "release red spoon"}, {"step": 7, "subtask": "grasp red spoon"}, {"step": 8, "subtask": "move grasped red spoon to the back left corner of the stove"}, {"step": 9, "subtask": "release red spoon"}, {"step": 10, "subtask": "grasp red spoon"}, {"step": 11, "subtask": "move grasped red spoon to the back right corner of the stove"}, {"step": 12, "subtask": "release red spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed39746680a92147fbfe4673.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0320/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 320, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 449, "role": "error_localization"}}
{"state": {"task": "remove the blueberries from the plate, then move the plate to the sink, then empty the plate into the sink", "plan": [{"step": 1, "subtask": "move grasped blueberries out"}, {"step": 2, "subtask": "release blueberries"}, {"step": 3, "subtask": "grasp plate"}, {"step": 4, "subtask": "move grasped plate to sink"}, {"step": 5, "subtask": "release plate"}, {"step": 6, "subtask": "push plate forward"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9239e1c25c73603f4f81ddff.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0321/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 321, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "0", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 307, "role": "error_localization"}}
{"state": {"task": "remove the chicken leg from the silver pot", "plan": [{"step": 1, "subtask": "release chicken leg"}, {"step": 2, "subtask": "move grasped silver pot upright"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "pick up chicken leg"}, {"step": 5, "subtask": "move grasped chicken leg away from silver pot"}, {"step": 6, "subtask": "pick up silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/605c2698db40b23da2e36801.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0322/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 322, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "14", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 298, "role": "error_localization"}}
{"state": {"task": "Move the green object into the silver pot, then out of the pot, then place it beside the pot, then back into the pot.", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object to silver pot"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object out of silver pot"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "grasp green object"}, {"step": 8, "subtask": "move grasped green object beside silver pot"}, {"step": 9, "subtask": "release green object"}, {"step": 10, "subtask": "grasp green object"}, {"step": 11, "subtask": "move grasped green object to silver pot"}, {"step": 12, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0323/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 323, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 424, "role": "error_localization"}}
{"state": {"task": "put the cup to the counter from the sink, then to the table, then back to the counter", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to table"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to counter"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e1eedd02af5c6c03881192b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0324/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/49", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 324, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "49", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 350, "role": "error_localization"}}
{"state": {"task": "move the blue spatula on the left burner, then move it to the right burner, then move it back to the left burner", "plan": [{"step": 1, "subtask": "grasp blue spatula"}, {"step": 2, "subtask": "move grasped blue spatula to right burner"}, {"step": 3, "subtask": "release blue spatula"}, {"step": 4, "subtask": "grasp blue spatula"}, {"step": 5, "subtask": "move grasped blue spatula to left burner"}, {"step": 6, "subtask": "release blue spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/783b3ec5d8c46e163ee8e7c4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0325/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 325, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 323, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the left top corner of the burner, then to the right top corner, then to the center of the burner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left top corner of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the right top corner of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center of the burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3414ba57fb6a578a86e319af.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0326/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/20", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 326, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "20", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 382, "role": "error_localization"}}
{"state": {"task": "take banana out of colander", "plan": [{"step": 1, "subtask": "move grasped banana away from colander"}, {"step": 2, "subtask": "move grasped colander to a stable surface"}, {"step": 3, "subtask": "release colander"}, {"step": 4, "subtask": "pick up banana"}, {"step": 5, "subtask": "pick up colander"}, {"step": 6, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/71968079ee95315f5cd44abe.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0327/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 327, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train", "episode_id": "16", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 296, "role": "error_localization"}}
{"state": {"task": "take the red object on the purple cloth, place it on the yellow box, then move it to the green tray, and finally return it to the purple cloth", "plan": [{"step": 1, "subtask": "grasp red object"}, {"step": 2, "subtask": "move grasped red object to yellow box"}, {"step": 3, "subtask": "release red object"}, {"step": 4, "subtask": "grasp red object"}, {"step": 5, "subtask": "move grasped red object to green tray"}, {"step": 6, "subtask": "release red object"}, {"step": 7, "subtask": "grasp red object"}, {"step": 8, "subtask": "move grasped red object to purple cloth"}, {"step": 9, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/187ca733a71978cd371599da.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0328/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_39_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 328, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_39_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 374, "role": "error_localization"}}
{"state": {"task": "move the yellow fish to the silver dish from the left stove", "plan": [{"step": 1, "subtask": "pick up yellow fish"}, {"step": 2, "subtask": "move grasped yellow fish to the silver dish"}, {"step": 3, "subtask": "release yellow fish"}, {"step": 4, "subtask": "release silver dish"}, {"step": 5, "subtask": "move grasped silver dish to the left stove"}, {"step": 6, "subtask": "pick up silver dish"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3c7bf5a9d12d373a77a77f1c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0329/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_27_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 329, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_27_train", "episode_id": "3", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 305, "role": "error_localization"}}
{"state": {"task": "remove the cucumber from the cup, then place it on the table, pick it up again, and put it back into the cup", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber out of cup"}, {"step": 3, "subtask": "release cucumber"}, {"step": 4, "subtask": "grasp cucumber"}, {"step": 5, "subtask": "move grasped cucumber to cup"}, {"step": 6, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ca455acb6df19780a4ec69db.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0330/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 330, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 310, "role": "error_localization"}}
{"state": {"task": "Put the yellow doll at the center of the table, then move it to the front left corner, then to the back right corner, finally return it to the center.", "plan": [{"step": 1, "subtask": "grasp yellow doll"}, {"step": 2, "subtask": "move grasped yellow doll to the center of the table"}, {"step": 3, "subtask": "release yellow doll"}, {"step": 4, "subtask": "grasp yellow doll"}, {"step": 5, "subtask": "move grasped yellow doll to the front left corner"}, {"step": 6, "subtask": "release yellow doll"}, {"step": 7, "subtask": "grasp yellow doll"}, {"step": 8, "subtask": "move grasped yellow doll to the back right corner"}, {"step": 9, "subtask": "release yellow doll"}, {"step": 10, "subtask": "grasp yellow doll"}, {"step": 11, "subtask": "move grasped yellow doll to the center of the table"}, {"step": 12, "subtask": "release yellow doll"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/223ec694abb51b4ba5e33564.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0331/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train/24", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 331, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train", "episode_id": "24", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 439, "role": "error_localization"}}
{"state": {"task": "Move the green toy to the bottom of the table, then to the top, then to the left side, then back to the bottom.", "plan": [{"step": 1, "subtask": "grasp green toy"}, {"step": 2, "subtask": "move grasped green toy to the bottom of the table"}, {"step": 3, "subtask": "release green toy"}, {"step": 4, "subtask": "grasp green toy"}, {"step": 5, "subtask": "move grasped green toy to the top of the table"}, {"step": 6, "subtask": "release green toy"}, {"step": 7, "subtask": "grasp green toy"}, {"step": 8, "subtask": "move grasped green toy to the left side of the table"}, {"step": 9, "subtask": "release green toy"}, {"step": 10, "subtask": "grasp green toy"}, {"step": 11, "subtask": "move grasped green toy to the bottom of the table"}, {"step": 12, "subtask": "release green toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b98ea03009591c137394fc15.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0332/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 332, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 436, "role": "error_localization"}}
{"state": {"task": "Remove cup from plate, then place it on the table, then return it to the plate, then move it to the tray.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to table"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to plate"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to tray"}, {"step": 12, "subtask": "return it to its original position"}, {"step": 13, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cf119044eef1f2d1c1f0beaf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0333/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/39", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 333, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "39", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 425, "role": "error_localization"}}
{"state": {"task": "Put the item closest to the sink into the sink", "plan": [{"step": 1, "subtask": "grasp potato"}, {"step": 2, "subtask": "move grasped potato to sink"}, {"step": 3, "subtask": "release potato"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0334/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 334, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 246, "role": "error_localization"}}
{"state": {"task": "close the drawer, open it, take out the pen, put it on the table, close the drawer again", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp drawer handle"}, {"step": 5, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 6, "subtask": "release drawer handle"}, {"step": 7, "subtask": "grasp pen"}, {"step": 8, "subtask": "move grasped pen out"}, {"step": 9, "subtask": "move grasped pen to table"}, {"step": 10, "subtask": "release drawer handle"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d54b05fdb7cf5b7311fd3735.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0335/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 335, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train", "episode_id": "6", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 383, "role": "error_localization"}}
{"state": {"task": "move the cloth to top left, then fold it neatly, then unfold it at the center of the table", "plan": [{"step": 1, "subtask": "release cloth"}, {"step": 2, "subtask": "grasp cloth"}, {"step": 3, "subtask": "move grasped cloth to center of table"}, {"step": 4, "subtask": "rotate grasped cloth"}, {"step": 5, "subtask": "push down"}, {"step": 6, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/674a937e2120c2e36734f7eb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0336/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train/37", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 336, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train", "episode_id": "37", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 303, "role": "error_localization"}}
{"state": {"task": "Open the pan lid, place the spoon inside, then remove the spoon, close the pan lid, and place the fork next to it.", "plan": [{"step": 1, "subtask": "grasp pan lid"}, {"step": 2, "subtask": "move grasped pan lid upwards"}, {"step": 3, "subtask": "release pan lid"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to inside pan"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon out of pan"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp pan lid"}, {"step": 11, "subtask": "move grasped pan lid downwards"}, {"step": 12, "subtask": "release pan lid"}, {"step": 13, "subtask": "grasp fork"}, {"step": 14, "subtask": "move grasped fork next to pan lid"}, {"step": 15, "subtask": "release fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/19ddcb9d3b683562792bb23e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0337/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 337, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 465, "role": "error_localization"}}
{"state": {"task": "put cup to drying rack from sink", "plan": [{"step": 1, "subtask": "pick up cup"}, {"step": 2, "subtask": "move grasped cup to countertop"}, {"step": 3, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/36d8c7acc6185f164efc472b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0338/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 338, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 244, "role": "error_localization"}}
{"state": {"task": "Move the can to the upper left side of the stove, then to the lower right side, then to the center of the stove", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the upper left side of the stove"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the lower right side of the stove"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the center of the stove"}, {"step": 9, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/af7c09cd6b203d424d53d8b5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0339/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 339, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "Put cup from counter to sink, then from sink to dish rack, then from dish rack to counter.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to sink"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to dish rack"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to counter"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/618f735c173259d70592d0ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0340/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/43", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 340, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "43", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "Take the lion toy and place it above the drawer, then move it beside the drawer, then place it on top of the drawer again.", "plan": [{"step": 1, "subtask": "grasp lion toy"}, {"step": 2, "subtask": "move grasped lion toy above the drawer"}, {"step": 3, "subtask": "release lion toy"}, {"step": 4, "subtask": "grasp lion toy"}, {"step": 5, "subtask": "move grasped lion toy beside the drawer"}, {"step": 6, "subtask": "release lion toy"}, {"step": 7, "subtask": "grasp lion toy"}, {"step": 8, "subtask": "move grasped lion toy to top of the drawer"}, {"step": 9, "subtask": "release lion toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/acd0111bd5e8882b17aef9c0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0341/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/41", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 341, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "41", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "move the cloth to the bottom, then fold it twice, then place it at the front corner, then unfold it halfway", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to bottom"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold grasped cloth"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "fold grasped cloth"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth to front corner"}, {"step": 12, "subtask": "release cloth"}, {"step": 13, "subtask": "grasp cloth"}, {"step": 14, "subtask": "unfold grasped cloth halfway"}, {"step": 15, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1af6cec5704137455abc579e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0342/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 342, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 450, "role": "error_localization"}}
{"state": {"task": "Move the blue towel to the upper left of burner, then to the lower right, then fold it neatly and place it next to the stove.", "plan": [{"step": 1, "subtask": "grasp blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the upper left of burner"}, {"step": 3, "subtask": "release blue towel"}, {"step": 4, "subtask": "grasp blue towel"}, {"step": 5, "subtask": "move grasped blue towel to the lower right"}, {"step": 6, "subtask": "release blue towel"}, {"step": 7, "subtask": "grasp blue towel"}, {"step": 8, "subtask": "fold blue towel"}, {"step": 9, "subtask": "move grasped blue towel next to the stove"}, {"step": 10, "subtask": "release blue towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5a1078e2668ddfbbb23fbbbc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0343/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 343, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 391, "role": "error_localization"}}
{"state": {"task": "Remove the cup from the plate, then place it on the table, then return it to the plate, then take it off again and place it on the plate upside down.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "release cup"}, {"step": 6, "subtask": "move grasped cup to table"}, {"step": 7, "subtask": "release cup"}, {"step": 8, "subtask": "grasp cup"}, {"step": 9, "subtask": "move grasped cup to plate"}, {"step": 10, "subtask": "release cup"}, {"step": 11, "subtask": "grasp cup"}, {"step": 12, "subtask": "rotate grasped cup"}, {"step": 13, "subtask": "move grasped cup to plate"}, {"step": 14, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/05e174d4bfb7f2c026e589f6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0344/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/30", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 344, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "30", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 447, "role": "error_localization"}}
{"state": {"task": "move the cloth from the top left to the centre of the table, then fold it, then unfold it, then place it at the bottom right", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the centre of the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth forward"}, {"step": 6, "subtask": "move grasped cloth to the bottom right"}, {"step": 7, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/48eab26340f105a65c972f11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0345/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_17_train/22", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 345, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_17_train", "episode_id": "22", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 335, "role": "error_localization"}}
{"state": {"task": "Remove blueberries from plate, place in bowl, then transfer to sink, then return to plate.", "plan": [{"step": 1, "subtask": "grasp blueberries"}, {"step": 2, "subtask": "move grasped blueberries to bowl"}, {"step": 3, "subtask": "put it back where it was"}, {"step": 4, "subtask": "release blueberries"}, {"step": 5, "subtask": "grasp blueberries"}, {"step": 6, "subtask": "move grasped blueberries to sink"}, {"step": 7, "subtask": "release blueberries"}, {"step": 8, "subtask": "grasp blueberries"}, {"step": 9, "subtask": "move grasped blueberries to plate"}, {"step": 10, "subtask": "release blueberries"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50178fd78a775935092228cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0346/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 346, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "13", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 377, "role": "error_localization"}}
{"state": {"task": "Move the yellow object to the pot from the left of the left side burner, then return it to its original position, then move it to the right of the right side burner.", "plan": [{"step": 1, "subtask": "grasp yellow object"}, {"step": 2, "subtask": "release yellow object"}, {"step": 3, "subtask": "grasp yellow object"}, {"step": 4, "subtask": "release yellow object"}, {"step": 5, "subtask": "grasp yellow object"}, {"step": 6, "subtask": "move grasped yellow object to the right of the right side burner"}, {"step": 7, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8eff56aaacb09f759a791e04.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0347/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 347, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "Move the yellow knife to the right bottom of the burner, then to the top left, then back to the right bottom, then to the center.", "plan": [{"step": 1, "subtask": "grasp yellow knife"}, {"step": 2, "subtask": "move grasped yellow knife to the right bottom of the burner"}, {"step": 3, "subtask": "release yellow knife"}, {"step": 4, "subtask": "grasp yellow knife"}, {"step": 5, "subtask": "move grasped yellow knife to the top left"}, {"step": 6, "subtask": "release yellow knife"}, {"step": 7, "subtask": "grasp yellow knife"}, {"step": 8, "subtask": "move grasped yellow knife to the right bottom of the burner"}, {"step": 9, "subtask": "release yellow knife"}, {"step": 10, "subtask": "grasp yellow knife"}, {"step": 11, "subtask": "move grasped yellow knife to the center"}, {"step": 12, "subtask": "release yellow knife"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/de3a0c877178726a880868d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0348/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 348, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_04_train", "episode_id": "16", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 434, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the right burner, then to the left burner, then back to the right burner", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the right burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the left burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the right burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f2a47d69693419ed1d256563.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0349/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 349, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "Move the blue spoon to the top left burner, then to the bottom right burner, then back to the top left burner.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the top left burner"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the bottom right burner"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to the top left burner"}, {"step": 9, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0e32deba42e08a4d5b70fb3e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0350/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 350, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "put the spoon in the bowl, then take it out, then put it in again", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon out"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to bowl"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8c0adb87a028b87e2cad7ecf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0351/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/29", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 351, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "29", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 346, "role": "error_localization"}}
{"state": {"task": "Put the spoon in the orange pot, then remove the lid from the pot, then put the spoon back in the pot.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to orange pot"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp pot lid"}, {"step": 5, "subtask": "move grasped pot lid away"}, {"step": 6, "subtask": "release pot lid"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to blue pot"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aed4303462eb9330157daa5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0352/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 352, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put the yellow can to the upper right corner", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth to upper right corner"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0353/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 353, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 427, "role": "error_localization"}}
{"state": {"task": "remove the green object from between the burners, then place it on the table, then return it to the original spot, then place it beside the stove", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object out from between the burners"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object to table"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "grasp green object"}, {"step": 8, "subtask": "move grasped green object to original spot"}, {"step": 9, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6ea37b21854ef0c98071472d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0354/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 354, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "15", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 375, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out the apple, place it on the table, then return it to the drawer, close the drawer.", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp apple"}, {"step": 5, "subtask": "move grasped apple out"}, {"step": 6, "subtask": "move grasped apple to table"}, {"step": 7, "subtask": "release apple"}, {"step": 8, "subtask": "grasp apple"}, {"step": 9, "subtask": "move grasped apple backward"}, {"step": 10, "subtask": "move grasped apple to drawer"}, {"step": 11, "subtask": "release apple"}, {"step": 12, "subtask": "grasp drawer handle"}, {"step": 13, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 14, "subtask": "release drawer handle"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9fbbca24a8a9fbb667d1f19c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0355/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 355, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 453, "role": "error_localization"}}
{"state": {"task": "Remove the wooden spatula from the basket, place it to the side of the right burner, then return it to the basket.", "plan": [{"step": 1, "subtask": "grasp wooden spatula"}, {"step": 2, "subtask": "move grasped wooden spatula out"}, {"step": 3, "subtask": "move grasped wooden spatula to the side of the right burner"}, {"step": 4, "subtask": "release wooden spatula"}, {"step": 5, "subtask": "grasp wooden spatula"}, {"step": 6, "subtask": "move grasped wooden spatula to basket"}, {"step": 7, "subtask": "release wooden spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ffb443ab28aa722706425e6e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0356/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_03_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 356, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_03_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "remove the green spatula from the right burner, then place it on the left burner, then return it to the right burner", "plan": [{"step": 1, "subtask": "grasp green spatula"}, {"step": 2, "subtask": "move grasped green spatula out of right burner"}, {"step": 3, "subtask": "release green spatula"}, {"step": 4, "subtask": "grasp green spatula"}, {"step": 5, "subtask": "move grasped green spatula to left burner"}, {"step": 6, "subtask": "release green spatula"}, {"step": 7, "subtask": "grasp green spatula"}, {"step": 8, "subtask": "release green spatula"}, {"step": 9, "subtask": "move grasped green spatula to right burner"}, {"step": 10, "subtask": "release green spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13acb27e9a90f1e76fe5bdf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0357/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_51_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 357, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_51_train", "episode_id": "16", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 394, "role": "error_localization"}}
{"state": {"task": "Put cup to drying rack from sink, then move plate from sink to drying rack, then fork from sink to drying rack", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to drying rack"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp plate"}, {"step": 5, "subtask": "move grasped plate to drying rack"}, {"step": 6, "subtask": "release plate"}, {"step": 7, "subtask": "grasp fork"}, {"step": 8, "subtask": "move grasped fork to drying rack"}, {"step": 9, "subtask": "release fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f0350fc1da496a4bb9b0fc6d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0358/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 358, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 357, "role": "error_localization"}}
{"state": {"task": "put into the silver pot the mushroom and the violet cloth, then put the silver pot in the sink", "plan": [{"step": 1, "subtask": "pick up mushroom"}, {"step": 2, "subtask": "move grasped mushroom to silver pot"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to pot"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to sink"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a11e82fa3b7b4b4505aa0993.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0359/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 359, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "move the spoon between the burners, then place it on the stove's back burner, then move it to the front burner", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon between the burners"}, {"step": 3, "subtask": "grasp spoon"}, {"step": 4, "subtask": "release spoon"}, {"step": 5, "subtask": "move grasped spoon to the stove's back burner"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to the front burner"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/624fcc2bfce1768bb84ae25f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0360/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 360, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train", "episode_id": "11", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 363, "role": "error_localization"}}
{"state": {"task": "Move the yellow cloth from the middle front of the burners to behind the right side of the burners, then to the left side, then back to the middle front.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to behind the right side of the burners"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth to the left side of the burners"}, {"step": 6, "subtask": "release yellow cloth"}, {"step": 7, "subtask": "grasp yellow cloth"}, {"step": 8, "subtask": "move grasped yellow cloth to the middle front of the burners"}, {"step": 9, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b13a7904f6d38c43dac100f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0361/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 361, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 392, "role": "error_localization"}}
{"state": {"task": "Move the corn to the stove, then place it into the pot on the stove, then move it back to the counter top.", "plan": [{"step": 1, "subtask": "grasp corn"}, {"step": 2, "subtask": "move grasped corn to the stove"}, {"step": 3, "subtask": "grasp corn"}, {"step": 4, "subtask": "grasp corn"}, {"step": 5, "subtask": "move grasped corn to the pot on the stove"}, {"step": 6, "subtask": "release corn"}, {"step": 7, "subtask": "release corn"}, {"step": 8, "subtask": "move grasped corn to the counter top"}, {"step": 9, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d52daf007d42b664fe0709bd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0362/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 362, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "2", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 362, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the orange cloth from the right side of the table, then place it on the green towel, then move it back to the orange cloth", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the orange cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "move grasped silver pot to the green towel"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to the orange cloth"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3928ce5b24db942b387b9579.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0363/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 363, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train", "episode_id": "13", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 393, "role": "error_localization"}}
{"state": {"task": "put spatula on plate sink, then move plate to drying rack, then move spatula to cutting board", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula to plate"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp plate"}, {"step": 5, "subtask": "move grasped plate to sink"}, {"step": 6, "subtask": "release plate"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to cutting board"}, {"step": 9, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f10a3cb9e75f3eacd9db109f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0364/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/40", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 364, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "40", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "Move the green spatula over the left burner, then move it to the right burner, then return it to the left burner, then move it to the center burner.", "plan": [{"step": 1, "subtask": "grasp green spatula"}, {"step": 2, "subtask": "move grasped green spatula to the left burner"}, {"step": 3, "subtask": "release green spatula"}, {"step": 4, "subtask": "grasp green spatula"}, {"step": 5, "subtask": "move grasped green spatula to the right burner"}, {"step": 6, "subtask": "release green spatula"}, {"step": 7, "subtask": "grasp green spatula"}, {"step": 8, "subtask": "move grasped green spatula to the left burner"}, {"step": 9, "subtask": "release green spatula"}, {"step": 10, "subtask": "grasp green spatula"}, {"step": 11, "subtask": "move grasped green spatula to the center burner"}, {"step": 12, "subtask": "release green spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9228b8eafda936aab2397303.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0365/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 365, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "Move the green spatula above the orange cloth, then below it, then to the left side of it, then back above it.", "plan": [{"step": 1, "subtask": "grasp green spatula"}, {"step": 2, "subtask": "move grasped green spatula above the orange cloth"}, {"step": 3, "subtask": "release green spatula"}, {"step": 4, "subtask": "grasp green spatula"}, {"step": 5, "subtask": "move grasped green spatula below the orange cloth"}, {"step": 6, "subtask": "release green spatula"}, {"step": 7, "subtask": "grasp green spatula"}, {"step": 8, "subtask": "move grasped green spatula to the left side of the orange cloth"}, {"step": 9, "subtask": "release green spatula"}, {"step": 10, "subtask": "grasp green spatula"}, {"step": 11, "subtask": "move grasped green spatula above the orange cloth"}, {"step": 12, "subtask": "release green spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a7eedc2106915f4f049ae6b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0366/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_83_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 366, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_83_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 442, "role": "error_localization"}}
{"state": {"task": "move the silver pot on the top of the left side burner of the stove, then to the center, then to the right side burner, then back to the left side burner", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the top of the left side burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the center of the stove"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the right side burner of the stove"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the top of the left side burner of the stove"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e172914fa52a8d6aca4c681.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0367/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 367, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 452, "role": "error_localization"}}
{"state": {"task": "Move the red object to the left burner, then to the right burner, then back to the left burner.", "plan": [{"step": 1, "subtask": "grasp red object"}, {"step": 2, "subtask": "move grasped red object to left burner"}, {"step": 3, "subtask": "release red object"}, {"step": 4, "subtask": "grasp red object"}, {"step": 5, "subtask": "release red object"}, {"step": 6, "subtask": "grasp red object"}, {"step": 7, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9d8611db6f5ba9488cf9e6b4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0368/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 368, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "17", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 323, "role": "error_localization"}}
{"state": {"task": "move blue spatula from right of stove to left burner, then to the pot, then to the pan, finally back to the stove", "plan": [{"step": 1, "subtask": "grasp blue spatula"}, {"step": 2, "subtask": "move grasped blue spatula to left burner"}, {"step": 3, "subtask": "release blue spatula"}, {"step": 4, "subtask": "grasp blue spatula"}, {"step": 5, "subtask": "move grasped blue spatula to pot"}, {"step": 6, "subtask": "release blue spatula"}, {"step": 7, "subtask": "grasp blue spatula"}, {"step": 8, "subtask": "move grasped blue spatula to pan"}, {"step": 9, "subtask": "release blue spatula"}, {"step": 10, "subtask": "grasp blue spatula"}, {"step": 11, "subtask": "move grasped blue spatula to stove"}, {"step": 12, "subtask": "release blue spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7921bf8c20b4ffe262d88957.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0369/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 369, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "put the blue fork on the orange towel, then put the red pepper in the pot, then put the pot in the sink", "plan": [{"step": 1, "subtask": "grasp blue fork"}, {"step": 2, "subtask": "move grasped blue fork to top of orange towel"}, {"step": 3, "subtask": "release blue fork"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to the pot"}, {"step": 6, "subtask": "release red pepper"}, {"step": 7, "subtask": "grasp red pepper"}, {"step": 8, "subtask": "move grasped object to the sink"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d361addd4233dde18166a4f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0370/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_107_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 370, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_107_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "take spoon out of bowl, then place it on the dish rack, then move it to the counter", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "move grasped spoon to dish rack"}, {"step": 4, "subtask": "release spoon"}, {"step": 5, "subtask": "grasp spoon"}, {"step": 6, "subtask": "move grasped spoon to counter"}, {"step": 7, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8910d4c69f20807a610ef6e3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0371/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/18", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 371, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "18", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 323, "role": "error_localization"}}
{"state": {"task": "Move the purple towel beside the blue cup, then under the cup, then on top of it", "plan": [{"step": 1, "subtask": "grasp purple towel"}, {"step": 2, "subtask": "move grasped purple towel beside the blue cup"}, {"step": 3, "subtask": "release purple towel"}, {"step": 4, "subtask": "grasp purple towel"}, {"step": 5, "subtask": "move grasped purple towel on top of the blue cup"}, {"step": 6, "subtask": "release purple towel"}, {"step": 7, "subtask": "grasp purple towel"}, {"step": 8, "subtask": "move grasped purple towel beside the blue cup"}, {"step": 9, "subtask": "release purple towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/64864b2765e3815eec8d1db6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0372/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 372, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the bottom left side of the stove, then the black pan to the center burner, and finally the silver pot to the top left side.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom left side of the stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp black pan"}, {"step": 5, "subtask": "move grasped black pan to the center burner"}, {"step": 6, "subtask": "release black pan"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the top left side of the stove"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc4e76630abe1a643f2a2624.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0373/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_80_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 373, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_80_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "Take the red fork and place it on the green cloth, then remove it and place it on the blue plate, then return it to the green cloth.", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to green cloth"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "grasp red fork"}, {"step": 5, "subtask": "move grasped red fork to blue plate"}, {"step": 6, "subtask": "release red fork"}, {"step": 7, "subtask": "grasp red fork"}, {"step": 8, "subtask": "move grasped red fork to green cloth"}, {"step": 9, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91cde32bc18d285d1f748739.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0374/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 374, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the left lower part of the burner, then return it to the right upper part, then to the middle of the burner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left lower part of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the right upper part of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the middle of the burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8bcef41ecfa18160b80f6d40.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0375/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_62_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 375, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_62_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 384, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the lower side of the right burner, then to the upper side, then back to the lower side, then to the center of the stove.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower side of the right burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper side of the right burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the lower side of the right burner"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the center of the stove"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac0da747917faa49255523b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0376/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_19_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 376, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_19_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 447, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the bottom left side of the stove, then the black pan to the center burner, and finally the silver pot to the top left side.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom left side of the stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped black pan to the center burner"}, {"step": 6, "subtask": "release black pan"}, {"step": 7, "subtask": "grasp black pan"}, {"step": 8, "subtask": "move grasped silver pot to the top left side of the stove"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc4e76630abe1a643f2a2624.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0377/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_80_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 377, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_80_train", "episode_id": "13", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "Move the yellow duck to the left side of the table, then to the right side, then back to the left, then place it behind the table", "plan": [{"step": 1, "subtask": "grasp yellow duck"}, {"step": 2, "subtask": "move grasped yellow duck to the left side of the table"}, {"step": 3, "subtask": "release yellow duck"}, {"step": 4, "subtask": "grasp yellow duck"}, {"step": 5, "subtask": "move grasped yellow duck to the right side of the table"}, {"step": 6, "subtask": "release yellow duck"}, {"step": 7, "subtask": "grasp yellow duck"}, {"step": 8, "subtask": "move grasped yellow duck to the left side of the table"}, {"step": 9, "subtask": "release yellow duck"}, {"step": 10, "subtask": "grasp yellow duck"}, {"step": 11, "subtask": "move grasped yellow duck behind the table"}, {"step": 12, "subtask": "release yellow duck"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bb8bfa7f21bf03ee2acad98b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0378/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 378, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_21_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 438, "role": "error_localization"}}
{"state": {"task": "grab the yellow brush and put it beneath the silver pot on the left burner, then the silver pot on the right burner, then return the yellow brush to its original place", "plan": [{"step": 1, "subtask": "grasp yellow brush"}, {"step": 2, "subtask": "move grasped yellow brush beneath the silver pot on the right burner"}, {"step": 3, "subtask": "release yellow brush"}, {"step": 4, "subtask": "grasp yellow brush"}, {"step": 5, "subtask": "move grasped yellow brush to its original place"}, {"step": 6, "subtask": "release yellow brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9d923360a78c057154f29dc5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0379/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_44_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 379, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_44_train", "episode_id": "17", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 331, "role": "error_localization"}}
{"state": {"task": "Remove the green object from the yellow cloth between the burners, then place it on the left burner, then move it to the right burner, then return it to the yellow cloth.", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object out of yellow cloth"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object to left burner"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "grasp green object"}, {"step": 8, "subtask": "move grasped green object to right burner"}, {"step": 9, "subtask": "release green object"}, {"step": 10, "subtask": "grasp green object"}, {"step": 11, "subtask": "move grasped green object to yellow cloth"}, {"step": 12, "subtask": "move it back to the original position"}, {"step": 13, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6ae788279934be2098cafe90.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0380/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 380, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train", "episode_id": "17", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 453, "role": "error_localization"}}
{"state": {"task": "move the blue towel to the bottom left of the stove, then the red towel from the drawer next to it, then the green towel from the box", "plan": [{"step": 1, "subtask": "grasp blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the top right of the stove"}, {"step": 3, "subtask": "release blue towel"}, {"step": 4, "subtask": "grasp drawer handle"}, {"step": 5, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 6, "subtask": "release drawer handle"}, {"step": 7, "subtask": "grasp red towel"}, {"step": 8, "subtask": "move grasped red towel out"}, {"step": 9, "subtask": "move grasped red towel to bottom left of the stove"}, {"step": 10, "subtask": "release red towel"}, {"step": 11, "subtask": "grasp green towel"}, {"step": 12, "subtask": "move grasped green towel out"}, {"step": 13, "subtask": "move grasped green towel to top right of the stove"}, {"step": 14, "subtask": "release green towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/983f41bd20a6a51e416e8cd0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0381/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 381, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 476, "role": "error_localization"}}
{"state": {"task": "put the cup to the counter from the sink, then to the stove, then back to the sink, then to the counter again", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to sink"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to counter"}, {"step": 12, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23e98788988f09c9c5829a59.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0382/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/29", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 382, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "29", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the edge of the table to the top, then fold it neatly, then place it on the chair.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the top of the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth forward"}, {"step": 6, "subtask": "move grasped cloth to the left"}, {"step": 7, "subtask": "move grasped cloth to the right"}, {"step": 8, "subtask": "release cloth"}, {"step": 9, "subtask": "grasp cloth"}, {"step": 10, "subtask": "move grasped cloth to the chair"}, {"step": 11, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0043486c552bff5e13c89b03.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0384/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 384, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 398, "role": "error_localization"}}
{"state": {"task": "pick up the banana and put it between the burners, then remove it and place it on the countertop, then pick it up again and return it between the burners", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana between the burners"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped banana to countertop"}, {"step": 6, "subtask": "release banana"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "release banana"}, {"step": 9, "subtask": "move grasped banana between the burners"}, {"step": 10, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/125b5342fba4fb88cb682373.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0385/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 385, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "8", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "Put the blue spoon to the left of the hot dog, then to its right, then above it, then below it.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the left of the hot dog"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the right of the hot dog"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon above the hot dog"}, {"step": 9, "subtask": "release blue spoon"}, {"step": 10, "subtask": "grasp blue spoon"}, {"step": 11, "subtask": "move grasped blue spoon below the hot dog"}, {"step": 12, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c41861b6f740f9b3288edc2b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0386/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 386, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 430, "role": "error_localization"}}
{"state": {"task": "remove the eggplant from the plate, then place it on the cutting board, then move it to the bowl", "plan": [{"step": 1, "subtask": "grasp eggplant"}, {"step": 2, "subtask": "move grasped eggplant out of plate"}, {"step": 3, "subtask": "move grasped eggplant to cutting board"}, {"step": 4, "subtask": "release eggplant"}, {"step": 5, "subtask": "grasp eggplant"}, {"step": 6, "subtask": "move grasped eggplant to bowl"}, {"step": 7, "subtask": "release eggplant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ede558588faf50ce8da7df16.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0387/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 387, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_eggplant_on_plate_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 332, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tap handle"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp bowl"}, {"step": 10, "subtask": "move grasped bowl to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp right burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner knob"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0388/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 388, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 580, "role": "error_localization"}}
{"state": {"task": "remove the knife from the top left from the table", "plan": [{"step": 1, "subtask": "pick up knife"}, {"step": 2, "subtask": "move grasped knife to the center of the table"}, {"step": 3, "subtask": "release knife"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c57f88c05cf053b13347a58a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0389/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 389, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 250, "role": "error_localization"}}
{"state": {"task": "Move the purple cloth to the upper right corner of the stove, then fold it neatly, then place it to the upper left corner.", "plan": [{"step": 1, "subtask": "grasp purple cloth"}, {"step": 2, "subtask": "move grasped purple cloth to the upper right corner of the stove"}, {"step": 3, "subtask": "release purple cloth"}, {"step": 4, "subtask": "grasp purple cloth"}, {"step": 5, "subtask": "rotate grasped purple cloth"}, {"step": 6, "subtask": "move grasped purple cloth to the upper left corner of the stove"}, {"step": 7, "subtask": "release purple cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/97a21ce502e0a35341e161cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0390/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 390, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "remove the blueberries from the plate, then move the plate to the sink, then empty the plate into the sink", "plan": [{"step": 1, "subtask": "grasp blueberries"}, {"step": 2, "subtask": "move grasped blueberries out"}, {"step": 3, "subtask": "release blueberries"}, {"step": 4, "subtask": "grasp plate"}, {"step": 5, "subtask": "move grasped plate to sink"}, {"step": 6, "subtask": "release plate"}, {"step": 7, "subtask": "push plate down"}, {"step": 8, "subtask": "push plate forward"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9239e1c25c73603f4f81ddff.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0391/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/0", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 391, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "0", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 338, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up bowl"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up corn"}, {"step": 5, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd8f638fc68e4fd658fe55cb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0392/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 392, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "19", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 270, "role": "error_localization"}}
{"state": {"task": "take the banana from the pan and put it on the right stove", "plan": [{"step": 1, "subtask": "pick up banana"}, {"step": 2, "subtask": "move grasped banana away from pan"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "pick up banana"}, {"step": 5, "subtask": "move grasped banana to right stove"}, {"step": 6, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/46a28e34a954dcf2f541797e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0393/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 393, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 298, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the left of burner, then to the right, then back to the left, then fold it under the burner", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth to the left of burner"}, {"step": 3, "subtask": "release blue cloth"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "move grasped blue cloth to the right"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp blue cloth"}, {"step": 8, "subtask": "move grasped blue cloth to the left of burner"}, {"step": 9, "subtask": "rotate grasped blue cloth"}, {"step": 10, "subtask": "push grasped blue cloth under burner"}, {"step": 11, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/094ab7bd16bc6b04557e54a0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0394/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 394, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 411, "role": "error_localization"}}
{"state": {"task": "move the red spoon to the front left corner of the stove, then to the front right corner, then to the back left corner, then to the back right corner", "plan": [{"step": 1, "subtask": "grasp red spoon"}, {"step": 2, "subtask": "move grasped red spoon to the front left corner of the stove"}, {"step": 3, "subtask": "release red spoon"}, {"step": 4, "subtask": "grasp red spoon"}, {"step": 5, "subtask": "move grasped red spoon to the front left corner of the stove"}, {"step": 6, "subtask": "release red spoon"}, {"step": 7, "subtask": "grasp red spoon"}, {"step": 8, "subtask": "move grasped red spoon to the back left corner of the stove"}, {"step": 9, "subtask": "release red spoon"}, {"step": 10, "subtask": "grasp red spoon"}, {"step": 11, "subtask": "move grasped red spoon to the back right corner of the stove"}, {"step": 12, "subtask": "release red spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed39746680a92147fbfe4673.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0395/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 395, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 449, "role": "error_localization"}}
{"state": {"task": "Put the blue spoon to the left of the hot dog, then to its right, then above it, then below it.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the left of the hot dog"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "release blue spoon"}, {"step": 6, "subtask": "move grasped blue spoon to the right of the hot dog"}, {"step": 7, "subtask": "release blue spoon"}, {"step": 8, "subtask": "grasp blue spoon"}, {"step": 9, "subtask": "move grasped blue spoon above the hot dog"}, {"step": 10, "subtask": "release blue spoon"}, {"step": 11, "subtask": "grasp blue spoon"}, {"step": 12, "subtask": "move grasped blue spoon below the hot dog"}, {"step": 13, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c41861b6f740f9b3288edc2b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0396/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 396, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "8", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 446, "role": "error_localization"}}
{"state": {"task": "Put the pot on top of the napkin, then place the pot on the table, then put it back on the napkin.", "plan": [{"step": 1, "subtask": "grasp pot"}, {"step": 2, "subtask": "move grasped pot to top of the napkin"}, {"step": 3, "subtask": "release pot"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "release pot"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to top of the napkin"}, {"step": 9, "subtask": "move grasped pot to table"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e5f6bd672ddccef0dda8196d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0397/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 397, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "16", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 364, "role": "error_localization"}}
{"state": {"task": "cook the red pepper in the pot with the oven for 5 minutes, put the pot on the stove, close the oven, then remove the red pepper from the pot, put it on the table, put the pot in the sink", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to pot"}, {"step": 6, "subtask": "release red pepper"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to oven"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp oven door handle"}, {"step": 11, "subtask": "move grasped door handle to close position"}, {"step": 12, "subtask": "release door handle"}, {"step": 13, "subtask": "wait 5 minutes"}, {"step": 14, "subtask": "grasp oven door handle"}, {"step": 15, "subtask": "move grasped door handle to open position"}, {"step": 16, "subtask": "release door handle"}, {"step": 17, "subtask": "grasp yellow plate"}, {"step": 18, "subtask": "move grasped yellow plate to stove"}, {"step": 19, "subtask": "release yellow plate"}, {"step": 20, "subtask": "grasp oven door handle"}, {"step": 21, "subtask": "move grasped door handle to close position"}, {"step": 22, "subtask": "release door handle"}, {"step": 23, "subtask": "grasp red pepper"}, {"step": 24, "subtask": "move grasped red pepper to table"}, {"step": 25, "subtask": "release red pepper"}, {"step": 26, "subtask": "grasp pot"}, {"step": 27, "subtask": "move grasped pot to sink"}, {"step": 28, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0398/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 398, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 723, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the bottom left of the burner, then to the top right, then back to the bottom left.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the top right of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top right of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the top left of the burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0bba7c89149c614833390126.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0399/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 399, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 378, "role": "error_localization"}}
{"state": {"task": "put the green cloth in the right side of the burner, then fold it neatly, then remove it and place it on the counter, then return it to the burner.", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth to the right side of the burner"}, {"step": 3, "subtask": "release green cloth"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "rotate grasped green cloth"}, {"step": 6, "subtask": "release green cloth"}, {"step": 7, "subtask": "grasp green cloth"}, {"step": 8, "subtask": "move grasped green cloth to the counter"}, {"step": 9, "subtask": "release green cloth"}, {"step": 10, "subtask": "grasp green cloth"}, {"step": 11, "subtask": "move grasped green cloth to the right side of the burner"}, {"step": 12, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5059793a7956e057cde276d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0400/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 400, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 434, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from the blue cloth, place it on the wooden table, then return it to the cloth, finally move it to the green mat.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out of blue cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "move grasped silver pot to wooden table"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "grasp silver pot"}, {"step": 7, "subtask": "move grasped silver pot to blue cloth"}, {"step": 8, "subtask": "release silver pot"}, {"step": 9, "subtask": "grasp silver pot"}, {"step": 10, "subtask": "move grasped silver pot to green mat"}, {"step": 11, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8ce9b9ac4ef5ac0b94b36f71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0401/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 401, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 410, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom left side of the burner, then to the top right side, then back to the bottom left side, then to the top left side", "plan": [{"step": 1, "subtask": "move grasped silver pot to the bottom left side of the burner"}, {"step": 2, "subtask": "release silver pot"}, {"step": 3, "subtask": "move grasped silver pot to the top right side of the burner"}, {"step": 4, "subtask": "release silver pot"}, {"step": 5, "subtask": "grasp silver pot"}, {"step": 6, "subtask": "move grasped silver pot to the bottom left side of the burner"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to the top left side of the burner"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0ddc5fe40decc5c40c65f500.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0402/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 402, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 416, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out each item one by one, place them on the table, then return them one by one into the drawer", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp object"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to table"}, {"step": 7, "subtask": "release object"}, {"step": 8, "subtask": "move grasped object to drawer"}, {"step": 9, "subtask": "move grasped object backward"}, {"step": 10, "subtask": "release object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a029113e8089645841bb6d8b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0403/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/26", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 403, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "26", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 383, "role": "error_localization"}}
{"state": {"task": "remove the cucumber from the cup, then place it on the table, pick it up again, and put it back into the cup", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "release cucumber"}, {"step": 3, "subtask": "move grasped cucumber out of cup"}, {"step": 4, "subtask": "release cucumber"}, {"step": 5, "subtask": "grasp cucumber"}, {"step": 6, "subtask": "move grasped cucumber to cup"}, {"step": 7, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ca455acb6df19780a4ec69db.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0404/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 404, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 324, "role": "error_localization"}}
{"state": {"task": "move the blue brush to upper side of the stove next to yellow cloth, then move it to lower side, then back to upper side", "plan": [{"step": 1, "subtask": "grasp blue brush"}, {"step": 2, "subtask": "move grasped blue brush to the lower side of the stove"}, {"step": 3, "subtask": "release blue brush"}, {"step": 4, "subtask": "grasp blue brush"}, {"step": 5, "subtask": "move grasped blue brush to the upper side of the stove"}, {"step": 6, "subtask": "release blue brush"}, {"step": 7, "subtask": "grasp blue brush"}, {"step": 8, "subtask": "move grasped blue brush to the lower side of the stove"}, {"step": 9, "subtask": "release blue brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0fbf5aaeb262929ab97488be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0405/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/16", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 405, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 381, "role": "error_localization"}}
{"state": {"task": "move the sushi to the left side of the table near the drawer, then to the right side, then back to the left side, then to the center", "plan": [{"step": 1, "subtask": "grasp sushi"}, {"step": 2, "subtask": "move grasped sushi to the left side of the table near the drawer"}, {"step": 3, "subtask": "release sushi"}, {"step": 4, "subtask": "grasp sushi"}, {"step": 5, "subtask": "move grasped sushi to the right side of the table"}, {"step": 6, "subtask": "release sushi"}, {"step": 7, "subtask": "grasp sushi"}, {"step": 8, "subtask": "move grasped sushi to the left side of the table near the drawer"}, {"step": 9, "subtask": "release sushi"}, {"step": 10, "subtask": "grasp sushi"}, {"step": 11, "subtask": "move grasped sushi to the center of the table"}, {"step": 12, "subtask": "release sushi"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/79ad9eb593086a1228047614.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0406/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_17_train/33", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 406, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_17_train", "episode_id": "33", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 436, "role": "error_localization"}}
{"state": {"task": "Move the silver pot on top of the green cloth, then place it on the blue mat, then return it to the cloth.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to blue mat"}, {"step": 6, "subtask": "bring it back to its original position"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dfd421c8d0c3cbba5e2cc99.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0407/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 407, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "3", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 393, "role": "error_localization"}}
{"state": {"task": "Move the white cloth from top edge of the burner to bottom edge, then back to top edge, then to bottom edge again.", "plan": [{"step": 1, "subtask": "grasp white cloth"}, {"step": 2, "subtask": "move grasped white cloth from top edge of the burner to bottom edge of the burner"}, {"step": 3, "subtask": "release white cloth"}, {"step": 4, "subtask": "grasp white cloth"}, {"step": 5, "subtask": "move grasped white cloth from bottom edge of the burner to top edge of the burner"}, {"step": 6, "subtask": "release white cloth"}, {"step": 7, "subtask": "grasp white cloth"}, {"step": 8, "subtask": "move grasped white cloth from top edge of the burner to bottom edge of the burner"}, {"step": 9, "subtask": "release white cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2702e87ff3dc7d7ea98d6801.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0408/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 408, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 394, "role": "error_localization"}}
{"state": {"task": "put cup from counter to sink, then return cup to counter, then put cup to sink again", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to sink"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to counter"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to sink"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9086571c2f6262666849c65.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0409/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 409, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 349, "role": "error_localization"}}
{"state": {"task": "Move the red pepper on top of the orange cloth, then on the blue plate, then back to the orange cloth.", "plan": [{"step": 1, "subtask": "grasp red pepper"}, {"step": 2, "subtask": "move grasped red pepper to top of the orange cloth"}, {"step": 3, "subtask": "release red pepper"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to blue plate"}, {"step": 6, "subtask": "release red pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7461e053a5b6e69086656776.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0410/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 410, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 317, "role": "error_localization"}}
{"state": {"task": "move the cloth from right to left, then fold it in half, then unfold it, then move it back to the right", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth from right to left"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold grasped cloth in half"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "unfold grasped cloth"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth to the right"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/db5d032dab6631088ee96ee0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0411/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 411, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_01_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "move the brush to the upper middle of the table, then to the upper left, then to the lower right", "plan": [{"step": 1, "subtask": "grasp brush"}, {"step": 2, "subtask": "move grasped brush to the upper middle of the table"}, {"step": 3, "subtask": "release brush"}, {"step": 4, "subtask": "grasp brush"}, {"step": 5, "subtask": "move grasped brush to the upper left of the table"}, {"step": 6, "subtask": "release brush"}, {"step": 7, "subtask": "grasp brush"}, {"step": 8, "subtask": "move grasped brush to the lower right of the table"}, {"step": 9, "subtask": "release brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7bc7090549ec933dfdb204cc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0412/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_15_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 412, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_15_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 367, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom right corner of the burner, then to the top left, then to the center", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom right corner of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the center of the burner"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c90750c2ccb5c44fcd5f9de5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0413/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 413, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 322, "role": "error_localization"}}
{"state": {"task": "Move the spoon to the left, then to the right, then to the top, then to the bottom.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to the left"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to the right"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to the top"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon to the bottom"}, {"step": 12, "subtask": "move it back to the original position"}, {"step": 13, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/58dc6b4bac1e4a0fa4b8d4fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0414/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_36_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 414, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_36_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 425, "role": "error_localization"}}
{"state": {"task": "Move one of the cans to the upper left corner of the left burner, then to the upper right corner, then to the lower left corner, and finally to the lower right corner.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the upper left corner of the left burner"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the lower left corner of the left burner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the lower right corner of the left burner"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "move grasped can to the upper right corner of the left burner"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c850847225497b2c7a0b7727.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0415/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 415, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 444, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom right of the burner", "plan": [{"step": 1, "subtask": "pick up silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom right of the burner"}, {"step": 3, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/112df704260b5e2cb3057c14.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0416/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_66_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 416, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_66_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 255, "role": "error_localization"}}
{"state": {"task": "put cup from counter to sink, then to microwave, then to fridge handle", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to sink"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to microwave"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to fridge handle"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0a57b7b0339419b130f4f705.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0417/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/29", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 417, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "29", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 346, "role": "error_localization"}}
{"state": {"task": "remove the green vegetable from the right side of the burner, then place it on the left side, then return it to the right side, then remove it again.", "plan": [{"step": 1, "subtask": "grasp green vegetable"}, {"step": 2, "subtask": "move grasped green vegetable out"}, {"step": 3, "subtask": "move grasped green vegetable to left side of the burner"}, {"step": 4, "subtask": "release green vegetable"}, {"step": 5, "subtask": "grasp green vegetable"}, {"step": 6, "subtask": "move grasped green vegetable out"}, {"step": 7, "subtask": "move grasped green vegetable to left side of the burner"}, {"step": 8, "subtask": "release green vegetable"}, {"step": 9, "subtask": "grasp green vegetable"}, {"step": 10, "subtask": "move grasped green vegetable out"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec5cb938ff4a1a745a46f539.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0418/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 418, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 400, "role": "error_localization"}}
{"state": {"task": "move the blue towel to the upper left of burner, then to the upper right of burner, then to the lower left of burner, then to the lower right of burner", "plan": [{"step": 1, "subtask": "grasp blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the upper left of burner"}, {"step": 3, "subtask": "release blue towel"}, {"step": 4, "subtask": "grasp blue towel"}, {"step": 5, "subtask": "move grasped blue towel to the upper right of burner"}, {"step": 6, "subtask": "release blue towel"}, {"step": 7, "subtask": "grasp blue towel"}, {"step": 8, "subtask": "move grasped blue towel to the lower left of burner"}, {"step": 9, "subtask": "release blue towel"}, {"step": 10, "subtask": "grasp blue towel"}, {"step": 11, "subtask": "move grasped blue towel to the lower right of burner"}, {"step": 12, "subtask": "release blue towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5a1078e2668ddfbbb23fbbbc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0419/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 419, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train", "episode_id": "11", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 442, "role": "error_localization"}}
{"state": {"task": "remove cup from plate", "plan": [{"step": 1, "subtask": "pick up cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "place cup back on plate"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a7a4e8599971151e571b8b49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0420/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/23", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 420, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "23", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 244, "role": "error_localization"}}
{"state": {"task": "move the red fork to the left side of the stove, then to the right side, then back to the left, then place it on top of the stove", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "grasp red fork"}, {"step": 5, "subtask": "release red fork"}, {"step": 6, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 7, "subtask": "release red fork"}, {"step": 8, "subtask": "grasp red fork"}, {"step": 9, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 10, "subtask": "release red fork"}, {"step": 11, "subtask": "grasp red fork"}, {"step": 12, "subtask": "move grasped red fork to top of the stove"}, {"step": 13, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9aec55453901a208bc241d7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0421/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_07_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 421, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_07_train", "episode_id": "11", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 458, "role": "error_localization"}}
{"state": {"task": "put red pot on the right burner, add the tomato and cheese into it, turn on the corresponding burner", "plan": [{"step": 1, "subtask": "grasp red tomato"}, {"step": 2, "subtask": "move grasped red tomato to right burner"}, {"step": 3, "subtask": "release red tomato"}, {"step": 4, "subtask": "grasp tomato"}, {"step": 5, "subtask": "move grasped tomato into red pot"}, {"step": 6, "subtask": "release tomato"}, {"step": 7, "subtask": "grasp cheese"}, {"step": 8, "subtask": "move grasped cheese into red pot"}, {"step": 9, "subtask": "release cheese"}, {"step": 10, "subtask": "grasp right burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0422/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 422, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 408, "role": "error_localization"}}
{"state": {"task": "remove spoon from bowl, then place it on the countertop, then back into the bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to countertop"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to bowl"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a2f559337cc33cca5a1f802e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0423/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 423, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 350, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the right bottom of the burner, then to the left top, then to the center, then back to the right bottom", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the right bottom of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the right bottom of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center of the burner"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the left top of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5c33dae017f4a15ba5e162f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0424/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_56_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 424, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_56_train", "episode_id": "14", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 440, "role": "error_localization"}}
{"state": {"task": "put the cans in the cupboard, put the bread on the table, and put the red towel, and the orange bowl in the sink", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to cupboard"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp another can"}, {"step": 8, "subtask": "move grasped can to cupboard"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp croissant"}, {"step": 11, "subtask": "move grasped croissant to table"}, {"step": 12, "subtask": "release croissant"}, {"step": 13, "subtask": "grasp red towel"}, {"step": 14, "subtask": "move grasped red towel to sink"}, {"step": 15, "subtask": "release red towel"}, {"step": 16, "subtask": "grasp orange bowl"}, {"step": 17, "subtask": "move grasped orange bowl to sink"}, {"step": 18, "subtask": "release orange bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0425/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 425, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 522, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the vegetable can", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp brocoli"}, {"step": 11, "subtask": "move grasped brocoli to upper right corner"}, {"step": 12, "subtask": "release brocoli"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0426/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 426, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 430, "role": "error_localization"}}
{"state": {"task": "put cup from counter to sink, then move it to counter, then to table, then back to sink", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to sink"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to counter"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to table"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to sink"}, {"step": 12, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc5ec89012aeb7227cfd52ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0427/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 427, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 401, "role": "error_localization"}}
{"state": {"task": "turn the cloth to top left from bottom right, then fold it twice, then place it under the object on the right", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top left"}, {"step": 3, "subtask": "rotate grasped cloth"}, {"step": 4, "subtask": "fold grasped cloth"}, {"step": 5, "subtask": "fold grasped cloth"}, {"step": 6, "subtask": "move grasped cloth under object on the right"}, {"step": 7, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50d7644d1676f5e85b858bff.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0428/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 428, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train", "episode_id": "6", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 330, "role": "error_localization"}}
{"state": {"task": "move the yellow towel on the upper left corner of the stove, then fold it and place it on the upper right corner, then unfold it and return it to the upper left corner", "plan": [{"step": 1, "subtask": "grasp yellow towel"}, {"step": 2, "subtask": "move grasped yellow towel to the upper right corner of the stove"}, {"step": 3, "subtask": "release yellow towel"}, {"step": 4, "subtask": "grasp yellow towel"}, {"step": 5, "subtask": "rotate grasped yellow towel"}, {"step": 6, "subtask": "move grasped yellow towel to the upper left corner of the stove"}, {"step": 7, "subtask": "release yellow towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52949f18fe35fc8c79927aad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0429/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_94_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 429, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_94_train", "episode_id": "12", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 354, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the vegetable", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp tomato"}, {"step": 11, "subtask": "move grasped tomato to upper right corner"}, {"step": 12, "subtask": "release tomato"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0430/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 430, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 426, "role": "error_localization"}}
{"state": {"task": "Move the blue object to the top right spot of the right burner, then to the center, then to the bottom left spot, then back to the top right spot.", "plan": [{"step": 1, "subtask": "grasp blue object"}, {"step": 2, "subtask": "move grasped blue object to the top right spot of the right burner"}, {"step": 3, "subtask": "release blue object"}, {"step": 4, "subtask": "grasp blue object"}, {"step": 5, "subtask": "move grasped blue object to the center"}, {"step": 6, "subtask": "release blue object"}, {"step": 7, "subtask": "grasp blue object"}, {"step": 8, "subtask": "grasp blue object"}, {"step": 9, "subtask": "release blue object"}, {"step": 10, "subtask": "move grasped blue object to the bottom left spot"}, {"step": 11, "subtask": "move grasped blue object to the top right spot of the right burner"}, {"step": 12, "subtask": "release blue object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4abc430ddbc71789b2a7473a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0431/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_46_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 431, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_46_train", "episode_id": "9", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 443, "role": "error_localization"}}
{"state": {"task": "move the blue cup to the bottom left of the right burner, then to the top left, then to the bottom right, then back to the center of the right burner", "plan": [{"step": 1, "subtask": "grasp blue cup"}, {"step": 2, "subtask": "move grasped blue cup to the top left of the right burner"}, {"step": 3, "subtask": "release blue cup"}, {"step": 4, "subtask": "grasp blue cup"}, {"step": 5, "subtask": "move grasped blue cup to the top left of the right burner"}, {"step": 6, "subtask": "release blue cup"}, {"step": 7, "subtask": "grasp blue cup"}, {"step": 8, "subtask": "move grasped blue cup to the bottom left of the right burner"}, {"step": 9, "subtask": "release blue cup"}, {"step": 10, "subtask": "grasp blue cup"}, {"step": 11, "subtask": "move grasped blue cup to the top left of the right burner"}, {"step": 12, "subtask": "release blue cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d43ef9cab7d5a642a7e06349.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0432/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_56_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 432, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_56_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 450, "role": "error_localization"}}
{"state": {"task": "close the drawer, then open it again, take out all visible objects, place each one back into the drawer, and close it.", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp drawer handle"}, {"step": 5, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 6, "subtask": "release drawer handle"}, {"step": 7, "subtask": "grasp object"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "release object"}, {"step": 10, "subtask": "grasp object"}, {"step": 11, "subtask": "move grasped object to drawer"}, {"step": 12, "subtask": "release object"}, {"step": 13, "subtask": "grasp drawer handle"}, {"step": 14, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 15, "subtask": "release drawer handle"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cc373be66bb13a2241110e5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0433/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_06_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 433, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_06_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 473, "role": "error_localization"}}
{"state": {"task": "Move the cloth to bottom from top, then to the left, then to the right, then back to the bottom.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to left"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to right"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to bottom"}, {"step": 9, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1af6cec5704137455abc579e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0434/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 434, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "13", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "Move the can to the lower right of burner, then to the upper left, then to the center, then back to the lower right.", "plan": [{"step": 1, "subtask": "move grasped can to the lower right of burner"}, {"step": 2, "subtask": "move grasped can to the lower right of burner"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the upper left"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the center"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "grasp can"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/295587a23223a9f08e2e105b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0435/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 435, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train", "episode_id": "1", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 418, "role": "error_localization"}}
{"state": {"task": "put the cup to the counter from the sink, then to the stove, then back to the sink, then to the counter again", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to sink"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to counter"}, {"step": 12, "subtask": "put it back where it was"}, {"step": 13, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23e98788988f09c9c5829a59.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0436/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/29", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 436, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "29", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 425, "role": "error_localization"}}
{"state": {"task": "remove the silver pot from top of the green cloth, then place it on the blue tray, then move the green cloth to the corner, then return the silver pot to its original position", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth to corner"}, {"step": 6, "subtask": "release green cloth"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to original position"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2e631394f8d985105fa1f975.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0437/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 437, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 376, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the upper right side of the stove, then to the upper left side, then to the lower right side, then back to the upper right side", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper right side of the stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper left side of the stove"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the lower right side of the stove"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the upper right side of the stove"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/117646b5fdd635f4290a7309.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0438/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 438, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 450, "role": "error_localization"}}
{"state": {"task": "move cloth to top from bottom, then fold it, then place it to the side.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "rotate grasped cloth"}, {"step": 6, "subtask": "push down"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "move grasped cloth to side"}, {"step": 10, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fbd5256059f152bb7404ce4b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0439/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 439, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "move the cloth from the left burner to the right burner", "plan": [{"step": 1, "subtask": "pick up cloth"}, {"step": 2, "subtask": "move grasped cloth to the left burner"}, {"step": 3, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0475b18118c518a6e66c64b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0440/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 440, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 249, "role": "error_localization"}}
{"state": {"task": "remove spoon from bowl, then place spoon on countertop, then put spoon back in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "release spoon"}, {"step": 6, "subtask": "move grasped spoon to countertop"}, {"step": 7, "subtask": "release spoon"}, {"step": 8, "subtask": "grasp spoon"}, {"step": 9, "subtask": "move grasped spoon to bowl"}, {"step": 10, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f9951d0fcde3b0635648233f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0441/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/28", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 441, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "28", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 365, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the left side of the burner, then to the front, then to the right side, and finally back to the left.", "plan": [{"step": 1, "subtask": "release silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left side of the burner"}, {"step": 3, "subtask": "grasp silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the front"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the right side of the burner"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the left side of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8f970e9600791f7a83997e2b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0442/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 442, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "13", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "take spatula off plate, place it on the counter, then back on the plate", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula out"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to plate"}, {"step": 6, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4410143e73acaefe1325e94c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0443/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spatula_off_plate_sink_train/5", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 443, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spatula_off_plate_sink_train", "episode_id": "5", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 305, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth to the bottom right corner of the table, then fold it, then unfold it, then move it to the top left corner.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to the bottom right corner of the table"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth forward"}, {"step": 6, "subtask": "move grasped yellow cloth to folded position"}, {"step": 7, "subtask": "release yellow cloth"}, {"step": 8, "subtask": "grasp yellow cloth"}, {"step": 9, "subtask": "move grasped yellow cloth from folded position"}, {"step": 10, "subtask": "move grasped yellow cloth to unfolded position"}, {"step": 11, "subtask": "release yellow cloth"}, {"step": 12, "subtask": "grasp yellow cloth"}, {"step": 13, "subtask": "move grasped yellow cloth to the top left corner of the table"}, {"step": 14, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ab4ea075c00e99b119946741.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0444/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_04_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 444, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_04_train", "episode_id": "13", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 475, "role": "error_localization"}}
{"state": {"task": "Move the blue cloth to the upper right of the table, then fold it neatly, then place it on the back left corner, then unfold it.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth to the upper right of the table"}, {"step": 3, "subtask": "release blue cloth"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "fold blue cloth neatly"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp blue cloth"}, {"step": 8, "subtask": "move grasped blue cloth to the back left corner"}, {"step": 9, "subtask": "release blue cloth"}, {"step": 10, "subtask": "grasp blue cloth"}, {"step": 11, "subtask": "unfold blue cloth"}, {"step": 12, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9cefbfdf0e942418f2953e94.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0445/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 445, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 422, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up bowl"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up corn"}, {"step": 5, "subtask": "move grasped corn to sink"}, {"step": 6, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd8f638fc68e4fd658fe55cb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0446/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 446, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "19", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 288, "role": "error_localization"}}
{"state": {"task": "Take the silver pot from the lower left side of the table to the upper center, then place it on the red plate, then move it back to the center, then lift it above the green bowl.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper center"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the red plate"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot up"}, {"step": 12, "subtask": "move grasped silver pot above the green bowl"}, {"step": 13, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5637b2725d4a5353bc7d991c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0447/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 447, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_29_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 458, "role": "error_localization"}}
{"state": {"task": "try to pull up the green cloth, then fold it, then put it inside the box", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth out"}, {"step": 3, "subtask": "rotate grasped green cloth"}, {"step": 4, "subtask": "move grasped green cloth to target"}, {"step": 5, "subtask": "release green cloth"}, {"step": 6, "subtask": "grasp green cloth"}, {"step": 7, "subtask": "move grasped green cloth to inside box"}, {"step": 8, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fabe07a6742a375eeff13925.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0448/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 448, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 343, "role": "error_localization"}}
{"state": {"task": "heat the orange in the microwave then put it in the pot, and put the silver lid on the pot, then put the pot on the bottom stove and turn it on", "plan": [{"step": 1, "subtask": "grasp orange"}, {"step": 2, "subtask": "move grasped orange to microwave"}, {"step": 3, "subtask": "release orange"}, {"step": 4, "subtask": "grasp orange"}, {"step": 5, "subtask": "move grasped orange out of microwave"}, {"step": 6, "subtask": "move grasped orange to the pot"}, {"step": 7, "subtask": "release orange"}, {"step": 8, "subtask": "grasp silver lid"}, {"step": 9, "subtask": "move silver lid to top of pot"}, {"step": 10, "subtask": "release silver lid"}, {"step": 11, "subtask": "grasp pot"}, {"step": 12, "subtask": "move grasped silver pot to the upper stove"}, {"step": 13, "subtask": "release pot"}, {"step": 14, "subtask": "grasp upper stove knob"}, {"step": 15, "subtask": "turn it on"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/113282c5a87d7796e1031a58.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0449/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 449, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 489, "role": "error_localization"}}
{"state": {"task": "place the pot on the top left of the kitchen burner, then move it to the top right, then to the bottom left, then to the bottom right", "plan": [{"step": 1, "subtask": "grasp pot"}, {"step": 2, "subtask": "move grasped pot to the top left of the kitchen burner"}, {"step": 3, "subtask": "release pot"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to the top right of the kitchen burner"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to the bottom left of the kitchen burner"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp pot"}, {"step": 11, "subtask": "move grasped pot to the bottom right of the kitchen burner"}, {"step": 12, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fb017213b23fdffb9443fd79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0450/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_41_train/2", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 450, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_41_train", "episode_id": "2", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 435, "role": "error_localization"}}
{"state": {"task": "Put the orange cloth on the left stove, then fold it neatly, move it to the right stove, then refold it, then return it to the left stove.", "plan": [{"step": 1, "subtask": "grasp orange cloth"}, {"step": 2, "subtask": "move grasped orange cloth to left stove"}, {"step": 3, "subtask": "release orange cloth"}, {"step": 4, "subtask": "grasp orange cloth"}, {"step": 5, "subtask": "move grasped orange cloth out"}, {"step": 6, "subtask": "rotate grasped orange cloth"}, {"step": 7, "subtask": "move grasped orange cloth to right stove"}, {"step": 8, "subtask": "release orange cloth"}, {"step": 9, "subtask": "grasp orange cloth"}, {"step": 10, "subtask": "move grasped orange cloth out"}, {"step": 11, "subtask": "rotate grasped orange cloth"}, {"step": 12, "subtask": "move grasped orange cloth to left stove"}, {"step": 13, "subtask": "release orange cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1952198e9c8e29d312d672f2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0451/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 451, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_45_train", "episode_id": "10", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 449, "role": "error_localization"}}
{"state": {"task": "move the yellow towel on the upper left corner of the stove, then fold it and place it on the upper right corner, then unfold it and return it to the upper left corner", "plan": [{"step": 1, "subtask": "grasp yellow towel"}, {"step": 2, "subtask": "move grasped yellow towel to the upper right corner of the stove"}, {"step": 3, "subtask": "release yellow towel"}, {"step": 4, "subtask": "grasp yellow towel"}, {"step": 5, "subtask": "release yellow towel"}, {"step": 6, "subtask": "move grasped yellow towel to the upper left corner of the stove"}, {"step": 7, "subtask": "rotate grasped yellow towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52949f18fe35fc8c79927aad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0452/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_94_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 452, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_94_train", "episode_id": "12", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 354, "role": "error_localization"}}
{"state": {"task": "remove cup from plate", "plan": [{"step": 1, "subtask": "pick up cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a7a4e8599971151e571b8b49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0453/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/23", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 453, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "23", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 241, "role": "error_localization"}}
{"state": {"task": "Put the yellow doll at the center of the table, then move it to the front left corner, then to the back right corner, finally return it to the center.", "plan": [{"step": 1, "subtask": "grasp yellow doll"}, {"step": 2, "subtask": "move grasped yellow doll to the center of the table"}, {"step": 3, "subtask": "release yellow doll"}, {"step": 4, "subtask": "grasp yellow doll"}, {"step": 5, "subtask": "move grasped yellow doll to the front left corner"}, {"step": 6, "subtask": "release yellow doll"}, {"step": 7, "subtask": "grasp yellow doll"}, {"step": 8, "subtask": "move grasped yellow doll to the front left corner"}, {"step": 9, "subtask": "release yellow doll"}, {"step": 10, "subtask": "grasp yellow doll"}, {"step": 11, "subtask": "move grasped yellow doll to the center of the table"}, {"step": 12, "subtask": "release yellow doll"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/223ec694abb51b4ba5e33564.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0454/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train/24", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 454, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train", "episode_id": "24", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 439, "role": "error_localization"}}
{"state": {"task": "Move the purple cloth to the bottom right corner of the stove, then fold it, then unfold it, then move it to the bottom left corner of the stove.", "plan": [{"step": 1, "subtask": "grasp purple cloth"}, {"step": 2, "subtask": "move grasped purple cloth to the bottom right corner of the stove"}, {"step": 3, "subtask": "release purple cloth"}, {"step": 4, "subtask": "grasp purple cloth"}, {"step": 5, "subtask": "move grasped purple cloth to folded position"}, {"step": 6, "subtask": "bring it back to its original position"}, {"step": 7, "subtask": "release purple cloth"}, {"step": 8, "subtask": "grasp purple cloth"}, {"step": 9, "subtask": "move grasped purple cloth to unfolded position"}, {"step": 10, "subtask": "release purple cloth"}, {"step": 11, "subtask": "grasp purple cloth"}, {"step": 12, "subtask": "move grasped purple cloth to the bottom left corner of the stove"}, {"step": 13, "subtask": "release purple cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/188043555a15ef77efd456e0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0455/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_106_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 455, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_106_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 458, "role": "error_localization"}}
{"state": {"task": "Move the yellow spatula to the right side of the burner, then to the front, then to the back, then to the left side.", "plan": [{"step": 1, "subtask": "grasp yellow spatula"}, {"step": 2, "subtask": "move grasped yellow spatula to the right side of the burner"}, {"step": 3, "subtask": "release yellow spatula"}, {"step": 4, "subtask": "grasp yellow spatula"}, {"step": 5, "subtask": "move grasped yellow spatula to the front"}, {"step": 6, "subtask": "release yellow spatula"}, {"step": 7, "subtask": "grasp yellow spatula"}, {"step": 8, "subtask": "move grasped yellow spatula to the back"}, {"step": 9, "subtask": "release yellow spatula"}, {"step": 10, "subtask": "grasp yellow spatula"}, {"step": 11, "subtask": "move grasped yellow spatula to the left side"}, {"step": 12, "subtask": "release yellow spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c42b354c174671d99573e52.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0456/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 456, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "9", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 441, "role": "error_localization"}}
{"state": {"task": "move bread to the left burner, then to the right burner, then to the center burner.", "plan": [{"step": 1, "subtask": "grasp bread"}, {"step": 2, "subtask": "move grasped bread to the left burner"}, {"step": 3, "subtask": "release bread"}, {"step": 4, "subtask": "grasp bread"}, {"step": 5, "subtask": "move grasped bread to the right burner"}, {"step": 6, "subtask": "release bread"}, {"step": 7, "subtask": "grasp bread"}, {"step": 8, "subtask": "move grasped bread to the center burner"}, {"step": 9, "subtask": "release bread"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3685251aafdb26adbde129d6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0457/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_15_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 457, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_15_train", "episode_id": "21", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 354, "role": "error_localization"}}
{"state": {"task": "Move the yellow banana between the silver pot and the blue cloth, then place it on the silver pot, then return it to the original position.", "plan": [{"step": 1, "subtask": "grasp yellow banana"}, {"step": 2, "subtask": "move grasped yellow banana between the silver pot and the blue cloth"}, {"step": 3, "subtask": "release yellow banana"}, {"step": 4, "subtask": "grasp yellow banana"}, {"step": 5, "subtask": "move grasped yellow banana to the silver pot"}, {"step": 6, "subtask": "release yellow banana"}, {"step": 7, "subtask": "grasp yellow banana"}, {"step": 8, "subtask": "move grasped yellow banana to original position"}, {"step": 9, "subtask": "release yellow banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/099b6feec710b502144f656f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0458/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 458, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "13", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 376, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from top of the green cloth, place it on the table, then return it to the cloth, finally lift the cloth to reveal what's underneath.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "grasp green cloth"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "move grasped silver pot to green cloth"}, {"step": 8, "subtask": "move grasped green cloth up"}, {"step": 9, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/63fcd9930f4272aa1004534d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0459/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 459, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the edge of the table to the top, then fold it neatly, then place it on the chair.", "plan": [{"step": 1, "subtask": "move grasped cloth to the top of the table"}, {"step": 2, "subtask": "grasp cloth"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth forward"}, {"step": 6, "subtask": "move grasped cloth to the left"}, {"step": 7, "subtask": "move grasped cloth to the right"}, {"step": 8, "subtask": "release cloth"}, {"step": 9, "subtask": "grasp cloth"}, {"step": 10, "subtask": "move grasped cloth to the chair"}, {"step": 11, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0043486c552bff5e13c89b03.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0460/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 460, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "4", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 398, "role": "error_localization"}}
{"state": {"task": "Take the red fork and place it on the green cloth, then remove it and place it on the blue plate, then return it to the green cloth.", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to green cloth"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "release red fork"}, {"step": 5, "subtask": "grasp red fork"}, {"step": 6, "subtask": "move grasped red fork to green cloth"}, {"step": 7, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91cde32bc18d285d1f748739.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0461/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 461, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 336, "role": "error_localization"}}
{"state": {"task": "move the towel from the bottom to the top", "plan": [{"step": 1, "subtask": "pick up towel"}, {"step": 2, "subtask": "move grasped towel to the top"}, {"step": 3, "subtask": "release towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7a5c68277f533188d0612980.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0462/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_43_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 462, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_43_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 246, "role": "error_localization"}}
{"state": {"task": "Put the cup from the counter to the sink, then return it to the counter, then move it to the fridge.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to sink"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to counter"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to fridge"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6a8fe056cc5a0cd4a151e15.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0463/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/57", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 463, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "57", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "Move the blue spoon to the right burner, then to the left burner, then back to the right burner.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the left burner"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the right burner"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to the right burner"}, {"step": 9, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6d56a60cccc6ceb77269f32.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0464/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 464, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train", "episode_id": "6", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 366, "role": "error_localization"}}
{"state": {"task": "put the red object on top of the orange cloth", "plan": [{"step": 1, "subtask": "pick up red object"}, {"step": 2, "subtask": "move grasped red object to top of the orange cloth"}, {"step": 3, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f7f2d6ef8aa1897c31a8dd0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0465/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 465, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 253, "role": "error_localization"}}
{"state": {"task": "move blue doll from top of the drawer and place it on the table", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up blue doll"}, {"step": 3, "subtask": "move grasped blue doll to drawer"}, {"step": 4, "subtask": "release blue doll"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ad6671a2052a36849b18cad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0466/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 466, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 267, "role": "error_localization"}}
{"state": {"task": "move the blue towel to the upper left of burner, then to the upper right of burner, then to the lower left of burner, then to the lower right of burner", "plan": [{"step": 1, "subtask": "grasp blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the upper left of burner"}, {"step": 3, "subtask": "release blue towel"}, {"step": 4, "subtask": "grasp blue towel"}, {"step": 5, "subtask": "move grasped blue towel to the lower left of burner"}, {"step": 6, "subtask": "release blue towel"}, {"step": 7, "subtask": "grasp blue towel"}, {"step": 8, "subtask": "move grasped blue towel to the lower left of burner"}, {"step": 9, "subtask": "release blue towel"}, {"step": 10, "subtask": "grasp blue towel"}, {"step": 11, "subtask": "move grasped blue towel to the lower right of burner"}, {"step": 12, "subtask": "release blue towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5a1078e2668ddfbbb23fbbbc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0467/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train/11", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 467, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 442, "role": "error_localization"}}
{"state": {"task": "put red pot on the right burner, add the tomato and cheese into it, turn on the corresponding burner", "plan": [{"step": 1, "subtask": "grasp red pot"}, {"step": 2, "subtask": "move grasped red pot to right burner"}, {"step": 3, "subtask": "release red pot"}, {"step": 4, "subtask": "grasp tomato"}, {"step": 5, "subtask": "move grasped tomato into red pot"}, {"step": 6, "subtask": "release tomato"}, {"step": 7, "subtask": "grasp yellow peach"}, {"step": 8, "subtask": "move grasped yellow peach into red pot"}, {"step": 9, "subtask": "release yellow peach"}, {"step": 10, "subtask": "grasp right burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0468/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 468, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 411, "role": "error_localization"}}
{"state": {"task": "place the blue fork below the right burner, then to the left burner, then back below the right burner", "plan": [{"step": 1, "subtask": "grasp blue fork"}, {"step": 2, "subtask": "move grasped blue fork below the right burner"}, {"step": 3, "subtask": "put it back where it was"}, {"step": 4, "subtask": "release blue fork"}, {"step": 5, "subtask": "grasp blue fork"}, {"step": 6, "subtask": "move grasped blue fork to the left burner"}, {"step": 7, "subtask": "release blue fork"}, {"step": 8, "subtask": "grasp blue fork"}, {"step": 9, "subtask": "move grasped blue fork below the right burner"}, {"step": 10, "subtask": "release blue fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3bdf5023e0932785300f5e01.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0469/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_107_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 469, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_107_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "put the cucumber in the cup, then remove the cucumber, then put it back in, then tilt the cup to roll out the cucumber", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber to cup"}, {"step": 3, "subtask": "release cucumber"}, {"step": 4, "subtask": "grasp cucumber"}, {"step": 5, "subtask": "move grasped cucumber out"}, {"step": 6, "subtask": "release cucumber"}, {"step": 7, "subtask": "grasp cucumber"}, {"step": 8, "subtask": "move grasped cucumber to table"}, {"step": 9, "subtask": "release cucumber"}, {"step": 10, "subtask": "grasp cucumber"}, {"step": 11, "subtask": "move grasped cucumber forward"}, {"step": 12, "subtask": "rotate grasped cucumber"}, {"step": 13, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/287cb4f6a2dca96741ac6d87.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0470/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/22", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 470, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "22", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 422, "role": "error_localization"}}
{"state": {"task": "Put cup to counter from sink, then to stove, then to table, then back to counter.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to table"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to counter"}, {"step": 12, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7f1df4e058567094aceee68f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0471/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 471, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 399, "role": "error_localization"}}
{"state": {"task": "move the white and orange object on the right burner", "plan": [{"step": 1, "subtask": "pick up white object"}, {"step": 2, "subtask": "move grasped white object to right burner"}, {"step": 3, "subtask": "release white object"}, {"step": 4, "subtask": "move grasped orange object to right burner"}, {"step": 5, "subtask": "release orange object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d2041f4c15a4c9033cbdba7d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0472/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train/12", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 472, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train", "episode_id": "12", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 285, "role": "error_localization"}}
{"state": {"task": "Remove the cucumber from the cup, then place the cucumber next to the cup, then return the cucumber to the cup, then remove the cucumber again.", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber out of cup"}, {"step": 3, "subtask": "release cucumber"}, {"step": 4, "subtask": "grasp cucumber"}, {"step": 5, "subtask": "move grasped cucumber to next to cup"}, {"step": 6, "subtask": "release cucumber"}, {"step": 7, "subtask": "grasp cucumber"}, {"step": 8, "subtask": "move grasped cucumber to cup"}, {"step": 9, "subtask": "release cucumber"}, {"step": 10, "subtask": "grasp cucumber"}, {"step": 11, "subtask": "move grasped cucumber out of cup"}, {"step": 12, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c11d23b620392eba3a9f546f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0473/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/22", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 473, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "22", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 413, "role": "error_localization"}}
{"state": {"task": "Place the capsicum near the ladle, then move the ladle to the center of the table, then place the capsicum next to the ladle again.", "plan": [{"step": 1, "subtask": "grasp capsicum"}, {"step": 2, "subtask": "move grasped capsicum near the ladle"}, {"step": 3, "subtask": "release capsicum"}, {"step": 4, "subtask": "grasp capsicum"}, {"step": 5, "subtask": "move grasped ladle to the center of the table"}, {"step": 6, "subtask": "release ladle"}, {"step": 7, "subtask": "grasp ladle"}, {"step": 8, "subtask": "move grasped capsicum next to the ladle"}, {"step": 9, "subtask": "release capsicum"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/96837ab6c4383dc63e004864.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0474/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_104_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 474, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_104_train", "episode_id": "10", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 380, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp oven handle"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp bowl"}, {"step": 10, "subtask": "move grasped bowl to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp left burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner knob"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0475/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 475, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 580, "role": "error_localization"}}
{"state": {"task": "Move the napkin to the left side of the blue fork, then to the right side of the blue fork, then to the center of the table", "plan": [{"step": 1, "subtask": "grasp napkin"}, {"step": 2, "subtask": "move grasped napkin to the left side of the blue fork"}, {"step": 3, "subtask": "release napkin"}, {"step": 4, "subtask": "grasp napkin"}, {"step": 5, "subtask": "move grasped napkin to the right side of the blue fork"}, {"step": 6, "subtask": "release napkin"}, {"step": 7, "subtask": "grasp napkin"}, {"step": 8, "subtask": "move grasped napkin to the center of the table"}, {"step": 9, "subtask": "release napkin"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/98d58ec9799f25aac67d0e37.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0476/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 476, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "move the blue towel from the bottom side of the table to the top side, then fold it neatly, then place it back at the bottom side", "plan": [{"step": 1, "subtask": "grasp blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the top side of the table"}, {"step": 3, "subtask": "release blue towel"}, {"step": 4, "subtask": "grasp blue towel"}, {"step": 5, "subtask": "rotate grasped blue towel"}, {"step": 6, "subtask": "push down"}, {"step": 7, "subtask": "release blue towel"}, {"step": 8, "subtask": "grasp blue towel"}, {"step": 9, "subtask": "move grasped blue towel to the bottom side of the table"}, {"step": 10, "subtask": "release blue towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2a9d69a61d88d2b53a05dc6d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0477/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 477, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 391, "role": "error_localization"}}
{"state": {"task": "remove corn from bowl, place corn on counter, return bowl to sink, then retrieve corn and place back in bowl", "plan": [{"step": 1, "subtask": "grasp corn"}, {"step": 2, "subtask": "move grasped corn out"}, {"step": 3, "subtask": "release corn"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl to sink"}, {"step": 6, "subtask": "release bowl"}, {"step": 7, "subtask": "grasp corn"}, {"step": 8, "subtask": "move grasped corn to bowl"}, {"step": 9, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bdf054964393383338b02ae6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0478/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train/43", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 478, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_corn_in_bowl_sink_train", "episode_id": "43", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up corn"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up bowl"}, {"step": 5, "subtask": "move grasped corn to sink"}, {"step": 6, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd8f638fc68e4fd658fe55cb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0479/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/19", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 479, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "19", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 288, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the left, then to the right, then to the center, then back to the left.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "grasp silver pot"}, {"step": 7, "subtask": "move grasped silver pot to the center"}, {"step": 8, "subtask": "release silver pot"}, {"step": 9, "subtask": "grasp silver pot"}, {"step": 10, "subtask": "move grasped silver pot to the left"}, {"step": 11, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2eafcc3f152844874a2521a5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0480/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/4", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 480, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "4", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 398, "role": "error_localization"}}
{"state": {"task": "move the silver pot at the top left corner of the burner, then to the top right corner, then to the bottom left corner, then to the bottom right corner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom left corner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the bottom right corner"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1dc6356b27c036dfe2829225.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0481/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/6", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 481, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "6", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 328, "role": "error_localization"}}
{"state": {"task": "place the yellow brush on the blue cloth, then move the blue cloth to the green tray, then place the yellow brush back on the blue cloth.", "plan": [{"step": 1, "subtask": "grasp yellow brush"}, {"step": 2, "subtask": "move grasped yellow brush to blue cloth"}, {"step": 3, "subtask": "release yellow brush"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "move grasped blue cloth to green tray"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp yellow brush"}, {"step": 8, "subtask": "move grasped yellow brush to blue cloth"}, {"step": 9, "subtask": "release yellow brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0ae8aa96ce4023993413e5e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0482/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_93_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 482, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_93_train", "episode_id": "8", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "Move the purple cloth on the upper right side of the burner, then fold it neatly, place it on the countertop, then return it to the original position.", "plan": [{"step": 1, "subtask": "grasp purple cloth"}, {"step": 2, "subtask": "move grasped purple cloth out"}, {"step": 3, "subtask": "rotate grasped purple cloth"}, {"step": 4, "subtask": "move grasped purple cloth to countertop"}, {"step": 5, "subtask": "release purple cloth"}, {"step": 6, "subtask": "grasp purple cloth"}, {"step": 7, "subtask": "move grasped purple cloth to upper right side of burner"}, {"step": 8, "subtask": "release purple cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ab3ff0b8fde824ba8acc0318.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0483/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_79_train/17", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 483, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_79_train", "episode_id": "17", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 361, "role": "error_localization"}}
{"state": {"task": "Remove the blue cloth from the top edge of the cloth, then place it on the bottom edge, then return it to the top edge, then set it aside.", "plan": [{"step": 1, "subtask": "move grasped blue cloth away from the top edge of the cloth"}, {"step": 2, "subtask": "release blue cloth"}, {"step": 3, "subtask": "move grasped blue cloth to the bottom edge of the cloth"}, {"step": 4, "subtask": "release blue cloth"}, {"step": 5, "subtask": "grasp blue cloth"}, {"step": 6, "subtask": "move grasped blue cloth to the top edge of the cloth"}, {"step": 7, "subtask": "release blue cloth"}, {"step": 8, "subtask": "grasp blue cloth"}, {"step": 9, "subtask": "move grasped blue cloth away from the top edge of the cloth"}, {"step": 10, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86116ee6494d87d8d320afb2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0484/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/8", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 484, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 412, "role": "error_localization"}}
{"state": {"task": "Move the mushroom to the lower right of the cooktop, then to the upper left, then back to the lower right, then to the center.", "plan": [{"step": 1, "subtask": "grasp mushroom"}, {"step": 2, "subtask": "move grasped mushroom to the lower right of the cooktop"}, {"step": 3, "subtask": "release mushroom"}, {"step": 4, "subtask": "move grasped mushroom to the upper left of the cooktop"}, {"step": 5, "subtask": "release mushroom"}, {"step": 6, "subtask": "grasp mushroom"}, {"step": 7, "subtask": "release mushroom"}, {"step": 8, "subtask": "grasp mushroom"}, {"step": 9, "subtask": "move grasped mushroom to the center of the cooktop"}, {"step": 10, "subtask": "release mushroom"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dac1307576affb964ab9ef8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0485/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_95_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 485, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_95_train", "episode_id": "1", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 391, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the bottom left of the burner, then to the top right, then back to the bottom left.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom left of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top right of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the bottom left of the burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0bba7c89149c614833390126.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0486/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/14", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 486, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "14", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 378, "role": "error_localization"}}
{"state": {"task": "move the cheese to the right of the blue brush, then to the left of the green cup, then back to the right of the blue brush.", "plan": [{"step": 1, "subtask": "grasp cheese"}, {"step": 2, "subtask": "move grasped cheese to the left of the green cup"}, {"step": 3, "subtask": "release cheese"}, {"step": 4, "subtask": "grasp cheese"}, {"step": 5, "subtask": "move grasped cheese to the right of the blue brush"}, {"step": 6, "subtask": "release cheese"}, {"step": 7, "subtask": "grasp cheese"}, {"step": 8, "subtask": "move grasped cheese to the left of the green cup"}, {"step": 9, "subtask": "release cheese"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8ecd81906d7736b2c835391.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0487/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_91_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 487, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_91_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 374, "role": "error_localization"}}
{"state": {"task": "remove the yellow duck from the drawer", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "move grasped yellow duck away from drawer"}, {"step": 3, "subtask": "release yellow duck"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cadf3573d63e2515c5e0acd8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0488/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/30", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 488, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "30", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 245, "role": "error_localization"}}
{"state": {"task": "remove the lid from the silver pot", "plan": [{"step": 1, "subtask": "pick up silver pot"}, {"step": 2, "subtask": "move grasped silver pot to a stable surface"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "pick up silver pot lid"}, {"step": 5, "subtask": "move grasped silver pot lid away from the pot"}, {"step": 6, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/96e99dc76bc0547361aa6f8a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0489/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 489, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_18_train", "episode_id": "1", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 303, "role": "error_localization"}}
{"state": {"task": "move cloth to top from bottom, then fold it, then place it to the side.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to side"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "rotate grasped cloth"}, {"step": 6, "subtask": "push down"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "move grasped cloth to top"}, {"step": 10, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fbd5256059f152bb7404ce4b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0490/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/1", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 490, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "Move the yellow spoon to the upper centre of the table, then to the lower left corner, then back to the upper centre, then to the upper right corner.", "plan": [{"step": 1, "subtask": "grasp yellow spoon"}, {"step": 2, "subtask": "move grasped yellow spoon to the upper centre of the table"}, {"step": 3, "subtask": "release yellow spoon"}, {"step": 4, "subtask": "grasp yellow spoon"}, {"step": 5, "subtask": "move grasped yellow spoon to the lower left corner of the table"}, {"step": 6, "subtask": "release yellow spoon"}, {"step": 7, "subtask": "grasp yellow spoon"}, {"step": 8, "subtask": "move grasped yellow spoon to the upper centre of the table"}, {"step": 9, "subtask": "release yellow spoon"}, {"step": 10, "subtask": "grasp yellow spoon"}, {"step": 11, "subtask": "release yellow spoon"}, {"step": 12, "subtask": "move grasped yellow spoon to the upper right corner of the table"}, {"step": 13, "subtask": "release yellow spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52960557c3d47f6f695b3e8e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0491/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train/20", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 491, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train", "episode_id": "20", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 462, "role": "error_localization"}}
{"state": {"task": "move the cloth to right from left, then fold it, then unfold it", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the right"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold cloth"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "unfold cloth"}, {"step": 9, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0492/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 492, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 338, "role": "error_localization"}}
{"state": {"task": "Pick up the knife and put it at the opposite corner of the stove, then move it back to the starting position, then move it to the middle of the stove.", "plan": [{"step": 1, "subtask": "grasp knife"}, {"step": 2, "subtask": "move grasped knife to the opposite corner of the stove"}, {"step": 3, "subtask": "release knife"}, {"step": 4, "subtask": "grasp knife"}, {"step": 5, "subtask": "move grasped knife back to the starting position"}, {"step": 6, "subtask": "release knife"}, {"step": 7, "subtask": "grasp knife"}, {"step": 8, "subtask": "move grasped knife to the middle of the stove"}, {"step": 9, "subtask": "release knife"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6c04c9a4660d662a4d09965b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0493/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/3", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 493, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "3", "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 375, "role": "error_localization"}}
{"state": {"task": "put spatula on plate sink, then move plate to drying rack, then move spatula to cutting board", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula to plate"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to cutting board"}, {"step": 6, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f10a3cb9e75f3eacd9db109f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0494/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/40", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 494, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "40", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 311, "role": "error_localization"}}
{"state": {"task": "put cup from drying rack to sink, then take another cup from the drying rack to the shelf, and then move it to the sink", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to shelf"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp another cup"}, {"step": 5, "subtask": "move grasped cup to sink"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to drying rack"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b669d05bfc254c40877b9b34.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0495/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/9", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 495, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 359, "role": "error_localization"}}
{"state": {"task": "put the pot on the blue cloth, then move it to the right side of the cloth, then to the left side, then back to the center", "plan": [{"step": 1, "subtask": "grasp pot"}, {"step": 2, "subtask": "move grasped pot to blue cloth"}, {"step": 3, "subtask": "release pot"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "release pot"}, {"step": 6, "subtask": "move grasped pot to right side of blue cloth"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to left side of blue cloth"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp pot"}, {"step": 11, "subtask": "move grasped pot to center of blue cloth"}, {"step": 12, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/18cfac129f28f5ff3deb2ab2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0496/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_68_train/10", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 496, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_68_train", "episode_id": "10", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 422, "role": "error_localization"}}
{"state": {"task": "Move the can to the lower side of the right burner, then to the upper side, then back to the lower side.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "release can"}, {"step": 3, "subtask": "move grasped can to the lower side of the right burner"}, {"step": 4, "subtask": "release can"}, {"step": 5, "subtask": "grasp can"}, {"step": 6, "subtask": "move grasped can to the upper side of the right burner"}, {"step": 7, "subtask": "release can"}, {"step": 8, "subtask": "grasp can"}, {"step": 9, "subtask": "move grasped can to the lower side of the right burner"}, {"step": 10, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0c8de888bf45a9546a6c42bb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0497/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/13", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 497, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "13", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "take the banana out of the drawer and put it on the right side of the table", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up banana"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "move grasped banana to the right side of the table"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9bb4d4184bab31fef9ebd10a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0498/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_11_train/15", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 498, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_11_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 272, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl, then take bowl out of sink onto the table, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "move grasped bowl to the table"}, {"step": 7, "subtask": "release bowl"}, {"step": 8, "subtask": "grasp cabinet door handle"}, {"step": 9, "subtask": "move grasped door handle to open position"}, {"step": 10, "subtask": "release door handle"}, {"step": 11, "subtask": "grasp spoon"}, {"step": 12, "subtask": "move grasped spoon to cabinet"}, {"step": 13, "subtask": "release spoon"}, {"step": 14, "subtask": "grasp fork"}, {"step": 15, "subtask": "move grasped object to bowl"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "bdv2fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0499/bdv2fail_plan_correct", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 499, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 482, "role": "error_localization"}}
{"state": {"task": "Move the banana to the upper middle of the table, then to the upper left, then to the upper right, then back to the upper middle.", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana to the upper middle of the table"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped banana to the upper left of the table"}, {"step": 6, "subtask": "release banana"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped banana to the upper right of the table"}, {"step": 9, "subtask": "release banana"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped banana to the upper middle of the table"}, {"step": 12, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2a019097dd9a28e6ed69d6f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0000/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train/18", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 0, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_23_train", "episode_id": "18", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 458, "role": "error_localization"}}
{"state": {"task": "put the red pepper in the pot, then put the spatula and yellow cloth in the sink, and turn on the left burner", "plan": [{"step": 1, "subtask": "grasp red pepper"}, {"step": 2, "subtask": "move grasped red pepper to pot"}, {"step": 3, "subtask": "release red pepper"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to sink"}, {"step": 6, "subtask": "release spatula"}, {"step": 7, "subtask": "grasp yellow plate"}, {"step": 8, "subtask": "move grasped yellow plate to sink"}, {"step": 9, "subtask": "release yellow plate"}, {"step": 10, "subtask": "grasp left burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0006/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 6, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "Remove the yellow scrubber from the green cloth, then place the scrubber on the table, then replace it back on the cloth.", "plan": [{"step": 1, "subtask": "grasp yellow scrubber"}, {"step": 2, "subtask": "move grasped yellow scrubber out of green cloth"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release yellow scrubber"}, {"step": 5, "subtask": "grasp yellow scrubber"}, {"step": 6, "subtask": "move grasped yellow scrubber to green cloth"}, {"step": 7, "subtask": "release yellow scrubber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1c99f5cb2750d538b55f158b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0007/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/6", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 7, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "6", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "Move the yellow spatula to the right side of the burner, then to the front, then to the back, then to the left side.", "plan": [{"step": 1, "subtask": "grasp yellow spatula"}, {"step": 2, "subtask": "move grasped yellow spatula to the right side of the burner"}, {"step": 3, "subtask": "release yellow spatula"}, {"step": 4, "subtask": "grasp yellow spatula"}, {"step": 5, "subtask": "move grasped yellow spatula to the back"}, {"step": 6, "subtask": "release yellow spatula"}, {"step": 7, "subtask": "grasp yellow spatula"}, {"step": 8, "subtask": "move grasped yellow spatula to the left side"}, {"step": 9, "subtask": "release yellow spatula"}, {"step": 10, "subtask": "grasp yellow spatula"}, {"step": 11, "subtask": "move grasped yellow spatula to the right side of the burner"}, {"step": 12, "subtask": "release yellow spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c42b354c174671d99573e52.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0009/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train/9", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 9, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_06_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 474, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the stove, then to the back of the stove, then to the front of the stove.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the front of the stove"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7ab2ec5c60b65199ea5eabb9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0012/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_25_train/12", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 12, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_25_train", "episode_id": "12", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 346, "role": "error_localization"}}
{"state": {"task": "move the green object in the pot above the right burner.", "plan": [{"step": 1, "subtask": "pick up green object"}, {"step": 2, "subtask": "release green object"}, {"step": 3, "subtask": "pick up pot"}, {"step": 4, "subtask": "move grasped pot above the right burner"}, {"step": 5, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd2bd32f33d4e6aa80411a08.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0014/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_86_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 14, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_86_train", "episode_id": "11", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 309, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl, then take bowl out of sink onto the table, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "move grasped bowl to the table"}, {"step": 7, "subtask": "release bowl"}, {"step": 8, "subtask": "grasp cabinet door handle"}, {"step": 9, "subtask": "move grasped door handle to open position"}, {"step": 10, "subtask": "release door handle"}, {"step": 11, "subtask": "grasp strawberry"}, {"step": 12, "subtask": "move grasped strawberry to oven"}, {"step": 13, "subtask": "release strawberry"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0019/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 19, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 462, "role": "error_localization"}}
{"state": {"task": "place the brush on the purple cloth, then on the green mat, then back on the purple cloth", "plan": [{"step": 1, "subtask": "grasp brush"}, {"step": 2, "subtask": "move grasped brush to purple cloth"}, {"step": 3, "subtask": "release brush"}, {"step": 4, "subtask": "grasp brush"}, {"step": 5, "subtask": "move grasped brush to green mat"}, {"step": 6, "subtask": "release brush"}, {"step": 7, "subtask": "grasp brush"}, {"step": 8, "subtask": "move grasped brush to purple cloth"}, {"step": 9, "subtask": "release brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8890b949807b6acc0245e76b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0021/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_30_train/3", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 21, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_30_train", "episode_id": "3", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 382, "role": "error_localization"}}
{"state": {"task": "Move the red fork to the right side of the stove, then to the left side, then back to the right side.", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "release red fork"}, {"step": 5, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 6, "subtask": "release red fork"}, {"step": 7, "subtask": "grasp red fork"}, {"step": 8, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 9, "subtask": "grasp red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0022/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 22, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "10", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 407, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tap handle"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp bowl"}, {"step": 10, "subtask": "move grasped bowl to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp left burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0025/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 25, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 608, "role": "error_localization"}}
{"state": {"task": "put the item closest to the center of the right burner into the pot", "plan": [{"step": 1, "subtask": "grasp green towel"}, {"step": 2, "subtask": "move grasped green towel into the pot"}, {"step": 3, "subtask": "release green towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0027/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 27, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 283, "role": "error_localization"}}
{"state": {"task": "remove the yellow object from th ecloth in", "plan": [{"step": 1, "subtask": "pick up yellow object"}, {"step": 2, "subtask": "move grasped yellow object away from cloth"}, {"step": 3, "subtask": "put it back where it was"}, {"step": 4, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/18ce7a1e0faa7922f55bf051.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0028/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 28, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 296, "role": "error_localization"}}
{"state": {"task": "put red pot on the right burner, add the tomato and cheese into it, turn on the corresponding burner", "plan": [{"step": 1, "subtask": "grasp red pot"}, {"step": 2, "subtask": "move grasped red pot to right burner"}, {"step": 3, "subtask": "release red pot"}, {"step": 4, "subtask": "grasp tomato"}, {"step": 5, "subtask": "move grasped tomato into red pot"}, {"step": 6, "subtask": "release tomato"}, {"step": 7, "subtask": "grasp cheese"}, {"step": 8, "subtask": "move grasped cheese into red pot"}, {"step": 9, "subtask": "release cheese"}, {"step": 10, "subtask": "grasp left burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0029/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 29, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "heat the bread for 25 minutes in the oven, then put it back in the orange bowl", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl to oven"}, {"step": 6, "subtask": "release bowl"}, {"step": 7, "subtask": "grasp oven door handle"}, {"step": 8, "subtask": "move grasped door handle to close position"}, {"step": 9, "subtask": "release door handle"}, {"step": 10, "subtask": "wait 25 minutes"}, {"step": 11, "subtask": "grasp door handle"}, {"step": 12, "subtask": "move grasped door handle to open position"}, {"step": 13, "subtask": "release door handle"}, {"step": 14, "subtask": "grasp bread"}, {"step": 15, "subtask": "move grasped bread to orange bowl"}, {"step": 16, "subtask": "release bread"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0032/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 32, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 512, "role": "error_localization"}}
{"state": {"task": "move the slice of orange on the side of the table, then move it to the center, then move it back to the side, then pick up another slice and place it next to the first", "plan": [{"step": 1, "subtask": "grasp orange slice"}, {"step": 2, "subtask": "move grasped orange slice to the side of the table"}, {"step": 3, "subtask": "release orange slice"}, {"step": 4, "subtask": "grasp orange slice"}, {"step": 5, "subtask": "move grasped orange slice to the center of the table"}, {"step": 6, "subtask": "release orange slice"}, {"step": 7, "subtask": "grasp another orange slice"}, {"step": 8, "subtask": "move grasped orange slice next to the first orange slice"}, {"step": 9, "subtask": "release orange slice"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/082fb7b21b913ced4428ff74.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0033/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/19", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 33, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "19", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 420, "role": "error_localization"}}
{"state": {"task": "remove the lid of the silver pot and put it on the left side of the can", "plan": [{"step": 1, "subtask": "pick up silver pot lid"}, {"step": 2, "subtask": "move grasped silver pot lid away from the pot"}, {"step": 3, "subtask": "release silver pot lid"}, {"step": 4, "subtask": "move grasped silver pot lid to the right side of the can"}, {"step": 5, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/20ce42d5c4f4e70e53a66b47.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0035/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_39_train/9", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 35, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_39_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 331, "role": "error_localization"}}
{"state": {"task": "Take the potato, move it to the right side of the stove, then to the left, then back to the right.", "plan": [{"step": 1, "subtask": "grasp potato"}, {"step": 2, "subtask": "move grasped potato to the right side of the stove"}, {"step": 3, "subtask": "release potato"}, {"step": 4, "subtask": "move grasped potato to the left side of the stove"}, {"step": 5, "subtask": "release potato"}, {"step": 6, "subtask": "grasp potato"}, {"step": 7, "subtask": "move grasped potato to the right side of the stove"}, {"step": 8, "subtask": "release potato"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0036/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 36, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 383, "role": "error_localization"}}
{"state": {"task": "move silver pot over orange cloth on the right stove, then the left stove, then back to the right stove", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot over orange cloth on the right stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "move grasped silver pot to the left stove"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot over orange cloth on the right stove"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/28ccd04f674bb1bce1a2ad4b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0039/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 39, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_30_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 418, "role": "error_localization"}}
{"state": {"task": "move the yellow cloth to the bottom right corner of the table, then fold it, then unfold it, then move it to the top left corner.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to the bottom right corner of the table"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth forward"}, {"step": 6, "subtask": "move grasped yellow cloth to folded position"}, {"step": 7, "subtask": "release yellow cloth"}, {"step": 8, "subtask": "grasp yellow cloth"}, {"step": 9, "subtask": "move grasped yellow cloth from folded position"}, {"step": 10, "subtask": "move grasped yellow cloth to folded position"}, {"step": 11, "subtask": "release yellow cloth"}, {"step": 12, "subtask": "grasp yellow cloth"}, {"step": 13, "subtask": "move grasped yellow cloth to the bottom right corner of the table"}, {"step": 14, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ab4ea075c00e99b119946741.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0047/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_04_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 47, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_04_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 504, "role": "error_localization"}}
{"state": {"task": "move the pot above the left burner.", "plan": [{"step": 1, "subtask": "pick up pot"}, {"step": 2, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dbc84c7a8bc41a802ad66bab.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0048/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_76_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 48, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_76_train", "episode_id": "15", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 254, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the lower left side of the burner, then to the lower right side, then back to the lower left side", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower left side of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "move grasped silver pot to the lower right side of the burner"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to the lower left side of the burner"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/88f022e072caaccb232ef323.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0049/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_28_train/0", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 49, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_28_train", "episode_id": "0", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 429, "role": "error_localization"}}
{"state": {"task": "Put spoon in bowl, then in sink, then back in bowl, then move bowl to the counter.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "release spoon"}, {"step": 3, "subtask": "move grasped spoon to bowl"}, {"step": 4, "subtask": "release spoon"}, {"step": 5, "subtask": "grasp spoon"}, {"step": 6, "subtask": "move grasped spoon to sink"}, {"step": 7, "subtask": "release spoon"}, {"step": 8, "subtask": "grasp spoon"}, {"step": 9, "subtask": "move grasped spoon to bowl"}, {"step": 10, "subtask": "release spoon"}, {"step": 11, "subtask": "grasp bowl"}, {"step": 12, "subtask": "move grasped bowl to counter"}, {"step": 13, "subtask": "release bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ad5a66ec02e008736204152d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0050/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/25", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 50, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "25", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 444, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the lower right of the burner, then to the upper left, then to the center, then back to the lower right", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower right of the burner"}, {"step": 3, "subtask": "move grasped silver pot to the center"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper left"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "release silver pot"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the lower right of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/472bc8f5862b505c2e5ce76f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0051/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_59_train/5", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 51, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_59_train", "episode_id": "5", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 463, "role": "error_localization"}}
{"state": {"task": "move the green object in front of the pot, then behind the pot, then back in front", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object in front of the pot"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object behind the pot"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/81bf1afcd54f4294f5d8c347.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0054/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_34_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 54, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_34_train", "episode_id": "15", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 356, "role": "error_localization"}}
{"state": {"task": "Move the yellow can to the bottom right of the table right to the silver pot, then to the top left, then back to the original spot, then under the silver pot.", "plan": [{"step": 1, "subtask": "grasp yellow can"}, {"step": 2, "subtask": "move grasped yellow can to the bottom right of the table right to the silver pot"}, {"step": 3, "subtask": "release yellow can"}, {"step": 4, "subtask": "grasp yellow can"}, {"step": 5, "subtask": "move grasped yellow can to the top left"}, {"step": 6, "subtask": "release yellow can"}, {"step": 7, "subtask": "grasp yellow can"}, {"step": 8, "subtask": "move grasped yellow can to the original spot"}, {"step": 9, "subtask": "release yellow can"}, {"step": 10, "subtask": "grasp yellow can"}, {"step": 11, "subtask": "move grasped yellow can under the silver pot"}, {"step": 12, "subtask": "put it back where it was"}, {"step": 13, "subtask": "release yellow can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8a85b906c16c89323f8a093e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0055/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 55, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 491, "role": "error_localization"}}
{"state": {"task": "take the silver lid off the silver pot and put it on the left side of the silver pot", "plan": [{"step": 1, "subtask": "pick up silver lid"}, {"step": 2, "subtask": "release silver lid"}, {"step": 3, "subtask": "move grasped silver lid away from silver pot"}, {"step": 4, "subtask": "release silver lid"}, {"step": 5, "subtask": "move grasped silver lid to the left side of the silver pot"}, {"step": 6, "subtask": "release silver lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8e07266de493f89fdd46818e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0057/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 57, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_35_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "pick up scissors and put into drawer", "plan": [{"step": 1, "subtask": "pick up drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "release scissors"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e965f9d8e06b966ff135b275.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0058/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train/38", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 58, "taskvar": "numpy_256_bridge_data_v1_berkeley_tool_chest_pick_up_scissors_and_put_into_drawer_train", "episode_id": "38", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 292, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from the orange cloth, place it on the table, then return it to the orange cloth, then move it to the green towel.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out of orange cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp orange towel"}, {"step": 8, "subtask": "move grasped silver pot to orange cloth"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the green towel"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0060/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 60, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "put the item closest to the sink into the sink", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the sink"}, {"step": 3, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0064/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 64, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 276, "role": "error_localization"}}
{"state": {"task": "Move the stainless pot to the upper right side of the burner, then to the lower left side, then back to the upper right side", "plan": [{"step": 1, "subtask": "grasp stainless pot"}, {"step": 2, "subtask": "grasp stainless pot"}, {"step": 3, "subtask": "release stainless pot"}, {"step": 4, "subtask": "move grasped stainless pot to the upper right side of the burner"}, {"step": 5, "subtask": "move grasped stainless pot to the lower left side"}, {"step": 6, "subtask": "release stainless pot"}, {"step": 7, "subtask": "grasp stainless pot"}, {"step": 8, "subtask": "move grasped stainless pot to the upper right side of the burner"}, {"step": 9, "subtask": "release stainless pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2ad53c280039412e29d56b31.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0065/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 65, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "7", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 410, "role": "error_localization"}}
{"state": {"task": "remove the violet cloth from the bottom right corner of the stove, then fold it neatly, place it on the countertop, then put it back onto the stove", "plan": [{"step": 1, "subtask": "grasp violet cloth"}, {"step": 2, "subtask": "move grasped violet cloth away from the bottom right corner of the stove"}, {"step": 3, "subtask": "release violet cloth"}, {"step": 4, "subtask": "grasp violet cloth"}, {"step": 5, "subtask": "fold grasped violet cloth"}, {"step": 6, "subtask": "move grasped violet cloth to the floor"}, {"step": 7, "subtask": "release violet cloth"}, {"step": 8, "subtask": "grasp violet cloth"}, {"step": 9, "subtask": "move grasped violet cloth to bottom right corner of the stove"}, {"step": 10, "subtask": "release violet cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1463de8f48b1d9c93229d0de.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0066/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_84_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 66, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_84_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 431, "role": "error_localization"}}
{"state": {"task": "Move the silver pot on the lower left corner side of the cooker, then to the upper right corner, then back to the lower left corner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper left corner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the lower right corner"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2c879f03ee5ab220de74086d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0071/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_77_train/9", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 71, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_77_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "put cucumber in cup, then take cucumber out of cup, then put cucumber in cup again", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "release cucumber"}, {"step": 3, "subtask": "grasp cucumber"}, {"step": 4, "subtask": "move grasped cucumber out of cup"}, {"step": 5, "subtask": "release cucumber"}, {"step": 6, "subtask": "grasp cucumber"}, {"step": 7, "subtask": "move grasped cucumber to cup"}, {"step": 8, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5fbbaf7e9267c641569720e1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0073/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 73, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "Move the silver bowl to the upper left side of the stove, then to the upper right side, then to the lower left side, then to the lower right side.", "plan": [{"step": 1, "subtask": "grasp silver bowl"}, {"step": 2, "subtask": "move grasped silver bowl to the upper left side of the stove"}, {"step": 3, "subtask": "release silver bowl"}, {"step": 4, "subtask": "grasp silver bowl"}, {"step": 5, "subtask": "move grasped silver bowl to the upper right side of the stove"}, {"step": 6, "subtask": "release silver bowl"}, {"step": 7, "subtask": "grasp silver bowl"}, {"step": 8, "subtask": "move grasped silver bowl to the lower left side of the stove"}, {"step": 9, "subtask": "release silver bowl"}, {"step": 10, "subtask": "grasp silver bowl"}, {"step": 11, "subtask": "move grasped silver bowl to the lower right side of the stove"}, {"step": 12, "subtask": "put it back where it was"}, {"step": 13, "subtask": "release silver bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7faa098ac776b0b560cb4ef6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0074/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_19_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 74, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_19_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 497, "role": "error_localization"}}
{"state": {"task": "Open the pan lid, place the spoon inside, then remove the spoon, close the pan lid, and place the fork next to it.", "plan": [{"step": 1, "subtask": "grasp pan lid"}, {"step": 2, "subtask": "move grasped pan lid upwards"}, {"step": 3, "subtask": "release pan lid"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to inside pan"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon out of pan"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp pan lid"}, {"step": 11, "subtask": "move grasped pan lid downwards"}, {"step": 12, "subtask": "release pan lid"}, {"step": 13, "subtask": "grasp fork"}, {"step": 14, "subtask": "move grasped fork next to pan lid"}, {"step": 15, "subtask": "move it back to the original position"}, {"step": 16, "subtask": "release fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/19ddcb9d3b683562792bb23e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0075/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train/5", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 75, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_26_train", "episode_id": "5", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 514, "role": "error_localization"}}
{"state": {"task": "Put the blue cloth on the top edge of the cloth, then on the bottom edge, then the left edge, then the right edge.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth to the top edge of the cloth"}, {"step": 3, "subtask": "release blue cloth"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "release blue cloth"}, {"step": 6, "subtask": "grasp blue cloth"}, {"step": 7, "subtask": "move grasped blue cloth to the left edge of the cloth"}, {"step": 8, "subtask": "release blue cloth"}, {"step": 9, "subtask": "grasp blue cloth"}, {"step": 10, "subtask": "move grasped blue cloth to the right edge of the cloth"}, {"step": 11, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c4ecd811048f34b0d2746cd4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0079/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 79, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 443, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the can that is the closest to the microwave", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp yellow can"}, {"step": 11, "subtask": "move grasped yellow can to upper right corner"}, {"step": 12, "subtask": "release yellow can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0080/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 80, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 465, "role": "error_localization"}}
{"state": {"task": "Put the cup from the counter to the sink, then to the dish rack, then back to the counter", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "release cup"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to dish rack"}, {"step": 6, "subtask": "move grasped cup to sink"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to counter"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/47fae74b78086735b8de3be0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0081/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/52", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 81, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "52", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 381, "role": "error_localization"}}
{"state": {"task": "put cup to counter from sink, then take another cup from the counter to the sink, then put the first cup back to the sink", "plan": [{"step": 1, "subtask": "move grasped cup to sink"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp another cup"}, {"step": 5, "subtask": "grasp cup"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to sink"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d025406977e108f8b7331708.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0082/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/70", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 82, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "70", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "remove cup from plate, place it on the table, then move the plate to the sink, then return the cup to the plate", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "return it to its original position"}, {"step": 4, "subtask": "release cup"}, {"step": 5, "subtask": "grasp cup"}, {"step": 6, "subtask": "move grasped cup to table"}, {"step": 7, "subtask": "release cup"}, {"step": 8, "subtask": "grasp plate"}, {"step": 9, "subtask": "move grasped plate to sink"}, {"step": 10, "subtask": "release plate"}, {"step": 11, "subtask": "grasp cup"}, {"step": 12, "subtask": "move grasped cup to plate"}, {"step": 13, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6b9f1ba9d9e66b13306a4f11.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0083/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/27", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 83, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "27", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 455, "role": "error_localization"}}
{"state": {"task": "remove beet from pot sink", "plan": [{"step": 1, "subtask": "pick up pot"}, {"step": 2, "subtask": "move grasped pot to sink"}, {"step": 3, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0084/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 84, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 270, "role": "error_localization"}}
{"state": {"task": "put the item closest to the right burner into the cupboard", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon into the cupboard"}, {"step": 6, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0086/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 86, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 332, "role": "error_localization"}}
{"state": {"task": "Open the lip of the silver pot, move to the right upper side, then tilt it slightly to the left, then return it to its original position.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "release silver pot"}, {"step": 3, "subtask": "move grasped silver pot to the right upper side"}, {"step": 4, "subtask": "rotate grasped silver pot slightly to the left"}, {"step": 5, "subtask": "move grasped silver pot to original position"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/21e8524b6b602ef4ee0ea89b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0087/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/9", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 87, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "9", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 357, "role": "error_localization"}}
{"state": {"task": "heat the bread for 25 minutes in the oven, then put it back in the orange bowl", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp bread"}, {"step": 5, "subtask": "move grasped bread to oven"}, {"step": 6, "subtask": "release bread"}, {"step": 7, "subtask": "grasp oven door handle"}, {"step": 8, "subtask": "move grasped door handle to close position"}, {"step": 9, "subtask": "release door handle"}, {"step": 10, "subtask": "wait 25 minutes"}, {"step": 11, "subtask": "grasp door handle"}, {"step": 12, "subtask": "move grasped door handle to open position"}, {"step": 13, "subtask": "release door handle"}, {"step": 14, "subtask": "grasp orange bowl"}, {"step": 15, "subtask": "move grasped bread to orange bowl"}, {"step": 16, "subtask": "release bread"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0096/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 96, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 513, "role": "error_localization"}}
{"state": {"task": "move the red fork to the right side of the stove, then to the left side, then back to the right, then to the center", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "grasp red fork"}, {"step": 5, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 6, "subtask": "release red fork"}, {"step": 7, "subtask": "grasp red fork"}, {"step": 8, "subtask": "move grasped red fork to the center of the stove"}, {"step": 9, "subtask": "release red fork"}, {"step": 10, "subtask": "grasp red fork"}, {"step": 11, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 12, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/39614107f837770db43ca768.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0097/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 97, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_32_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 468, "role": "error_localization"}}
{"state": {"task": "put the pink cup and pink plate on the dryer, then put the carrot, blue cup, purple eggplant on the table", "plan": [{"step": 1, "subtask": "grasp pink cup"}, {"step": 2, "subtask": "move grasped pink cup to dryer"}, {"step": 3, "subtask": "release pink cup"}, {"step": 4, "subtask": "grasp pink plate"}, {"step": 5, "subtask": "move grasped pink plate to dryer"}, {"step": 6, "subtask": "release pink plate"}, {"step": 7, "subtask": "grasp carrot"}, {"step": 8, "subtask": "move grasped carrot to table"}, {"step": 9, "subtask": "release carrot"}, {"step": 10, "subtask": "grasp pink cup"}, {"step": 11, "subtask": "move grasped pink cup to table"}, {"step": 12, "subtask": "release pink cup"}, {"step": 13, "subtask": "grasp purple eggplant"}, {"step": 14, "subtask": "move grasped purple eggplant to table"}, {"step": 15, "subtask": "release purple eggplant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e1eedd02af5c6c03881192b1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0098/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/49", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 98, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "49", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 499, "role": "error_localization"}}
{"state": {"task": "Move the silver pot lid to the top right corner of the stove, then to the top left corner, then to the bottom right corner, then to the bottom left corner.", "plan": [{"step": 1, "subtask": "grasp silver pot lid"}, {"step": 2, "subtask": "move grasped silver pot lid to the top right corner of the stove"}, {"step": 3, "subtask": "release silver pot lid"}, {"step": 4, "subtask": "grasp silver pot lid"}, {"step": 5, "subtask": "move grasped silver pot lid to the bottom left corner of the stove"}, {"step": 6, "subtask": "release silver pot lid"}, {"step": 7, "subtask": "grasp silver pot lid"}, {"step": 8, "subtask": "move grasped silver pot lid to the bottom right corner of the stove"}, {"step": 9, "subtask": "release silver pot lid"}, {"step": 10, "subtask": "grasp silver pot lid"}, {"step": 11, "subtask": "move grasped silver pot lid to the top left corner of the stove"}, {"step": 12, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a6edae47522f8a97521f5736.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0099/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 99, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_35_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 491, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom of the table, then to the top, then to the middle, then to the bottom again", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom of the table"}, {"step": 3, "subtask": "grasp silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top of the table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the middle of the table"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "release silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the bottom of the table"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a628c2c636d373ca727d8859.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0100/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train/12", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 100, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_32_train", "episode_id": "12", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 463, "role": "error_localization"}}
{"state": {"task": "move the spatula on top of the orange towel, then move it to the green napkin, then back to the towel", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula to top of the orange towel"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to the green napkin"}, {"step": 6, "subtask": "release spatula"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to top of the orange towel"}, {"step": 9, "subtask": "bring it back to its original position"}, {"step": 10, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4b4a11467601363b8f736b90.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0101/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train/19", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 101, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_36_train", "episode_id": "19", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 424, "role": "error_localization"}}
{"state": {"task": "Move the purple eggplant to the right side of the right burner, then to the left side of the right burner, then to the center of the right burner, then to the back side of the right burner.", "plan": [{"step": 1, "subtask": "grasp purple eggplant"}, {"step": 2, "subtask": "release purple eggplant"}, {"step": 3, "subtask": "release purple eggplant"}, {"step": 4, "subtask": "grasp purple eggplant"}, {"step": 5, "subtask": "move grasped purple eggplant to the left side of the right burner"}, {"step": 6, "subtask": "release purple eggplant"}, {"step": 7, "subtask": "grasp purple eggplant"}, {"step": 8, "subtask": "move grasped purple eggplant to the center of the right burner"}, {"step": 9, "subtask": "move grasped purple eggplant to the right side of the right burner"}, {"step": 10, "subtask": "grasp purple eggplant"}, {"step": 11, "subtask": "move grasped purple eggplant to the back side of the right burner"}, {"step": 12, "subtask": "release purple eggplant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1f17290cb276461f65ac6253.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0105/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_99_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 105, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_99_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 498, "role": "error_localization"}}
{"state": {"task": "put the orange object inside the silver pot, take it out, then put it back in, then tilt the pot slightly to check stability", "plan": [{"step": 1, "subtask": "grasp orange object"}, {"step": 2, "subtask": "move grasped orange object to silver pot"}, {"step": 3, "subtask": "return it to its original position"}, {"step": 4, "subtask": "release orange object"}, {"step": 5, "subtask": "grasp orange object"}, {"step": 6, "subtask": "move grasped orange object out of silver pot"}, {"step": 7, "subtask": "release orange object"}, {"step": 8, "subtask": "grasp orange object"}, {"step": 9, "subtask": "move grasped orange object to silver pot"}, {"step": 10, "subtask": "release orange object"}, {"step": 11, "subtask": "grasp silver pot"}, {"step": 12, "subtask": "tilt grasped silver pot slightly"}, {"step": 13, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/429d4f6bfcb3d55c22a06982.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0107/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/18", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 107, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "18", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 471, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out the spoon, place it on the table, then take out the fork and place it next to the spoon", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon out"}, {"step": 6, "subtask": "move grasped spoon to table"}, {"step": 7, "subtask": "release spoon"}, {"step": 8, "subtask": "grasp fork"}, {"step": 9, "subtask": "move grasped fork to next to spoon"}, {"step": 10, "subtask": "release fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b8d957c9dcce3f3e9dd8a509.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0110/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train/40", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 110, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_24_train", "episode_id": "40", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 412, "role": "error_localization"}}
{"state": {"task": "put the red object on top of the orange cloth", "plan": [{"step": 1, "subtask": "move grasped red object to top of the orange cloth"}, {"step": 2, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4f7f2d6ef8aa1897c31a8dd0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0116/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 116, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 266, "role": "error_localization"}}
{"state": {"task": "put the item closest to the right burner into the cupboard", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper into the cupboard"}, {"step": 6, "subtask": "release red pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0117/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 117, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 332, "role": "error_localization"}}
{"state": {"task": "put the red pepper in the pot, then put the spatula and yellow cloth in the sink, and turn on the left burner", "plan": [{"step": 1, "subtask": "grasp red pepper"}, {"step": 2, "subtask": "move grasped red pepper to pot"}, {"step": 3, "subtask": "release red pepper"}, {"step": 4, "subtask": "grasp fork"}, {"step": 5, "subtask": "move grasped object to sink"}, {"step": 6, "subtask": "release object"}, {"step": 7, "subtask": "grasp yellow cloth"}, {"step": 8, "subtask": "move grasped yellow cloth to sink"}, {"step": 9, "subtask": "release yellow cloth"}, {"step": 10, "subtask": "grasp left burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0118/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 118, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 442, "role": "error_localization"}}
{"state": {"task": "Move the lid of the orange pot to the top left corner of the table, then to the bottom right, then back to the top left.", "plan": [{"step": 1, "subtask": "grasp orange pot lid"}, {"step": 2, "subtask": "move grasped orange pot lid to the top left corner of the table"}, {"step": 3, "subtask": "release orange pot lid"}, {"step": 4, "subtask": "grasp orange pot lid"}, {"step": 5, "subtask": "move grasped orange pot lid to the bottom right corner of the table"}, {"step": 6, "subtask": "release orange pot lid"}, {"step": 7, "subtask": "grasp orange pot lid"}, {"step": 8, "subtask": "release orange pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1fd2ee149fa09289b75e7469.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0119/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train/18", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 119, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_14_train", "episode_id": "18", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 397, "role": "error_localization"}}
{"state": {"task": "Put cup from counter to sink, then from sink to dish rack, then from dish rack to counter.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "release cup"}, {"step": 3, "subtask": "move grasped cup to sink"}, {"step": 4, "subtask": "release cup"}, {"step": 5, "subtask": "grasp cup"}, {"step": 6, "subtask": "move grasped cup to dish rack"}, {"step": 7, "subtask": "release cup"}, {"step": 8, "subtask": "grasp cup"}, {"step": 9, "subtask": "move grasped cup to counter"}, {"step": 10, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/618f735c173259d70592d0ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0120/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/43", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 120, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "43", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 395, "role": "error_localization"}}
{"state": {"task": "move the blue cloth to the left of burner, then to the right, then back to the left, then fold it under the burner", "plan": [{"step": 1, "subtask": "move grasped blue cloth to the left of burner"}, {"step": 2, "subtask": "release blue cloth"}, {"step": 3, "subtask": "grasp blue cloth"}, {"step": 4, "subtask": "move grasped blue cloth to the right"}, {"step": 5, "subtask": "release blue cloth"}, {"step": 6, "subtask": "move grasped blue cloth to the left of burner"}, {"step": 7, "subtask": "rotate grasped blue cloth"}, {"step": 8, "subtask": "push grasped blue cloth under burner"}, {"step": 9, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/094ab7bd16bc6b04557e54a0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0123/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_53_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 123, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_53_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "move the carrot to the drawer", "plan": [{"step": 1, "subtask": "pick up carrot"}, {"step": 2, "subtask": "open drawer"}, {"step": 3, "subtask": "move grasped carrot to drawer"}, {"step": 4, "subtask": "release carrot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc288b2bbc62dbcfc71c150c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0127/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_26_train/5", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 127, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_26_train", "episode_id": "5", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 285, "role": "error_localization"}}
{"state": {"task": "place the blue brush on top of the purple cloth", "plan": [{"step": 1, "subtask": "release blue brush"}, {"step": 2, "subtask": "move grasped blue brush to top of the purple cloth"}, {"step": 3, "subtask": "pick up blue brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c812d12fe8c925f72f94b05e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0130/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_79_train/16", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 130, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_79_train", "episode_id": "16", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 282, "role": "error_localization"}}
{"state": {"task": "put the cans in the cupboard, put the bread on the table, and put the red towel, and the orange bowl in the sink", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to cupboard"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp another can"}, {"step": 8, "subtask": "move grasped can to cupboard"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped bread to table"}, {"step": 12, "subtask": "release bread"}, {"step": 13, "subtask": "grasp red towel"}, {"step": 14, "subtask": "move grasped red towel to sink"}, {"step": 15, "subtask": "release red towel"}, {"step": 16, "subtask": "grasp orange"}, {"step": 17, "subtask": "move grasped orange to sink"}, {"step": 18, "subtask": "release orange"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0131/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 131, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 545, "role": "error_localization"}}
{"state": {"task": "place the blue towel at the front of the left burner", "plan": [{"step": 1, "subtask": "pick up blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the front of the left burner"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release blue towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fb63104f378cf38a3f4c526a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0132/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_88_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 132, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_88_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 303, "role": "error_localization"}}
{"state": {"task": "Remove the slice green bell pepper from the orange cloth, then place it on the table, then return it to the orange cloth", "plan": [{"step": 1, "subtask": "grasp green bell pepper"}, {"step": 2, "subtask": "move grasped green bell pepper out of the orange cloth"}, {"step": 3, "subtask": "release green bell pepper"}, {"step": 4, "subtask": "grasp green bell pepper"}, {"step": 5, "subtask": "move grasped green bell pepper to the orange cloth"}, {"step": 6, "subtask": "release green bell pepper"}, {"step": 7, "subtask": "grasp green bell pepper"}, {"step": 8, "subtask": "move grasped green bell pepper to the table"}, {"step": 9, "subtask": "release green bell pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4faf6548055fb452f6835661.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0133/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_41_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 133, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_41_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 408, "role": "error_localization"}}
{"state": {"task": "put the tallest object to the right of the right burner", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the right of the right burner"}, {"step": 3, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0135/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 135, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 284, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the upper left of the table, then to the lower right, then back to the upper left, then to the center.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper left of the table"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the lower right of the table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the upper left of the table"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/411c69c79a0e1233cc485d4c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0138/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 138, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "14", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 412, "role": "error_localization"}}
{"state": {"task": "Move the spoon to the bottom right corner of the burner, then to the top left corner, then to the center of the burner, then back to the bottom right corner.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to the bottom right corner of the burner"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to the center of the burner"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to the top left corner of the burner"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon to the bottom right corner of the burner"}, {"step": 12, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3f1e7392f8ded1be8e1c0536.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0139/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 139, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 465, "role": "error_localization"}}
{"state": {"task": "take spoon out of bowl, then place spoon in sink, then take spoon out of sink, then put spoon back in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to sink"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "move grasped spoon to bowl"}, {"step": 8, "subtask": "move grasped spoon out of sink"}, {"step": 9, "subtask": "grasp spoon"}, {"step": 10, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0140/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 140, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 405, "role": "error_localization"}}
{"state": {"task": "Place the yellow cloth to the upper right side of the burner, then fold it neatly, then unfold it fully, then tuck the corners under the burner.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to the lower left side of the burner"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth to fold it neatly"}, {"step": 6, "subtask": "release yellow cloth"}, {"step": 7, "subtask": "grasp yellow cloth"}, {"step": 8, "subtask": "move grasped yellow cloth to keep it folded"}, {"step": 9, "subtask": "release yellow cloth"}, {"step": 10, "subtask": "grasp yellow cloth"}, {"step": 11, "subtask": "move grasped yellow cloth to tuck the corners under the burner"}, {"step": 12, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/78eede20c59f1c760c4da3ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0141/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_47_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 141, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_47_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 468, "role": "error_localization"}}
{"state": {"task": "move the cloth from the left bottom side of the cooker to the top right side, then fold it, then move it to the left top side, then unfold it", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the top right side of the cooker"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "rotate grasped cloth"}, {"step": 6, "subtask": "push down"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth from right to left"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a440eaafa8ed44241d69ebf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0144/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 144, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "11", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 444, "role": "error_localization"}}
{"state": {"task": "move the cloth to the left, then to the center, then to the right, then back to the left", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "grasp cloth"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to the center"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to the right"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "move grasped cloth to the left"}, {"step": 11, "subtask": "move grasped cloth to the left"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/986e630af04cfad7046679d8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0145/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 145, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_08_train", "episode_id": "7", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 435, "role": "error_localization"}}
{"state": {"task": "take the blue cloth from the right side of the burner to the left, then to the front, then to the back, then fold it and place it on the counter.", "plan": [{"step": 1, "subtask": "grasp blue cloth"}, {"step": 2, "subtask": "move grasped blue cloth to the left"}, {"step": 3, "subtask": "move grasped blue cloth to the back"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "move grasped blue cloth to the front"}, {"step": 6, "subtask": "release blue cloth"}, {"step": 7, "subtask": "grasp blue cloth"}, {"step": 8, "subtask": "release blue cloth"}, {"step": 9, "subtask": "release blue cloth"}, {"step": 10, "subtask": "grasp blue cloth"}, {"step": 11, "subtask": "fold grasped blue cloth"}, {"step": 12, "subtask": "move grasped blue cloth to the counter"}, {"step": 13, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50b596b2e642d75615e7fccd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0146/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_82_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 146, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_82_train", "episode_id": "14", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 477, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the lower side of the right burner, then to the upper side, then back to the lower side, then to the center of the stove.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower side of the right burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper side of the right burner"}, {"step": 6, "subtask": "move grasped silver pot to the lower side of the right burner"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "release silver pot"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the center of the stove"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac0da747917faa49255523b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0147/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_19_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 147, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_19_train", "episode_id": "10", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 476, "role": "error_localization"}}
{"state": {"task": "Move the green spatula from the green cloth to center of the stove, then to the front left burner, then to the front right burner, then back to the center.", "plan": [{"step": 1, "subtask": "grasp green spatula"}, {"step": 2, "subtask": "move grasped green spatula to center of the stove"}, {"step": 3, "subtask": "release green spatula"}, {"step": 4, "subtask": "grasp green spatula"}, {"step": 5, "subtask": "move grasped green spatula to front left burner"}, {"step": 6, "subtask": "release green spatula"}, {"step": 7, "subtask": "grasp green spatula"}, {"step": 8, "subtask": "move grasped green spatula to front right burner"}, {"step": 9, "subtask": "put it back where it was"}, {"step": 10, "subtask": "release green spatula"}, {"step": 11, "subtask": "grasp green spatula"}, {"step": 12, "subtask": "move grasped green spatula to center of the stove"}, {"step": 13, "subtask": "release green spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/26fc795ac507115bdda01405.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0149/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 149, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_34_train", "episode_id": "2", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 496, "role": "error_localization"}}
{"state": {"task": "relocate the green object to the upper right side of the drawer", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "move grasped green object to the upper right side of the drawer"}, {"step": 3, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5bb33868f20c6b1f0335082c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0154/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/23", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 154, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "23", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 285, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the can that is the closest to the cyan towel", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp yellow can"}, {"step": 11, "subtask": "move grasped yellow can to upper right corner"}, {"step": 12, "subtask": "release yellow can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0156/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 156, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 466, "role": "error_localization"}}
{"state": {"task": "Put up the silver pot on the right burner, then move it to the left burner, then return it to the right burner.", "plan": [{"step": 1, "subtask": "move grasped silver pot to the left burner"}, {"step": 2, "subtask": "release silver pot"}, {"step": 3, "subtask": "move grasped silver pot to the right burner"}, {"step": 4, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14c5a0d430c195739fc480ee.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0162/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_06_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 162, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_06_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 315, "role": "error_localization"}}
{"state": {"task": "move the green cloth to the upper right corner of the stove, then to the center burner, then to the lower left corner, then fold it neatly over the handle", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth to the upper right corner of the stove"}, {"step": 3, "subtask": "release green cloth"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth to the lower left corner"}, {"step": 6, "subtask": "release green cloth"}, {"step": 7, "subtask": "grasp green cloth"}, {"step": 8, "subtask": "move grasped green cloth to the center burner"}, {"step": 9, "subtask": "release green cloth"}, {"step": 10, "subtask": "grasp green cloth"}, {"step": 11, "subtask": "move grasped green cloth over the handle"}, {"step": 12, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13fcbb5e08966d462633de16.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0163/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 163, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 466, "role": "error_localization"}}
{"state": {"task": "Put the green cloth on the right side of the burner, then fold it neatly, then unfold it back.", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth to the right side of the burner"}, {"step": 3, "subtask": "release green cloth"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth forward"}, {"step": 6, "subtask": "move grasped green cloth backward"}, {"step": 7, "subtask": "release green cloth"}, {"step": 8, "subtask": "grasp green cloth"}, {"step": 9, "subtask": "move grasped green cloth from right to left"}, {"step": 10, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c5059793a7956e057cde276d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0164/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 164, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "4", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 414, "role": "error_localization"}}
{"state": {"task": "Put the item closest to the left burner into the pot", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped object to pot"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0168/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 168, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 276, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out the apple, place it on the table, then return it to the drawer, close the drawer.", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp apple"}, {"step": 5, "subtask": "move grasped apple out"}, {"step": 6, "subtask": "move grasped apple to floor"}, {"step": 7, "subtask": "release apple"}, {"step": 8, "subtask": "grasp apple"}, {"step": 9, "subtask": "move grasped apple backward"}, {"step": 10, "subtask": "move grasped apple to drawer"}, {"step": 11, "subtask": "release apple"}, {"step": 12, "subtask": "grasp drawer handle"}, {"step": 13, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 14, "subtask": "release drawer handle"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9fbbca24a8a9fbb667d1f19c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0169/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 169, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_19_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 482, "role": "error_localization"}}
{"state": {"task": "put the yellow corn in the pot, put the pot on the right burner and turn it on", "plan": [{"step": 1, "subtask": "grasp yellow corn"}, {"step": 2, "subtask": "move grasped yellow corn to the pot"}, {"step": 3, "subtask": "release yellow corn"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to the right burner"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp left burner knob"}, {"step": 8, "subtask": "rotate grasped object to on position"}, {"step": 9, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ac62285427547839b7682924.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0170/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 170, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_02_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "put into the cupboard the pot, spatula, red pepper, yellow towel, remove from the cupboard a fork and spatula and put them to the table", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to cupboard"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to cupboard"}, {"step": 9, "subtask": "release spatula"}, {"step": 10, "subtask": "grasp red pepper"}, {"step": 11, "subtask": "move grasped red pepper to cupboard"}, {"step": 12, "subtask": "release red pepper"}, {"step": 13, "subtask": "grasp yellow towel"}, {"step": 14, "subtask": "move grasped yellow towel to cupboard"}, {"step": 15, "subtask": "release yellow towel"}, {"step": 16, "subtask": "grasp fork"}, {"step": 17, "subtask": "move grasped fork out of cupboard"}, {"step": 18, "subtask": "move grasped fork to table"}, {"step": 19, "subtask": "release fork"}, {"step": 20, "subtask": "grasp spatula"}, {"step": 21, "subtask": "move grasped spatula out of cupboard"}, {"step": 22, "subtask": "move grasped spatula to table"}, {"step": 23, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0173/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 173, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 648, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out the pen, place it on the table, then take out the book, place it next to the pen, then take out the key and place it on top of the book", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp pen"}, {"step": 5, "subtask": "move grasped pen out"}, {"step": 6, "subtask": "move grasped pen to drawer"}, {"step": 7, "subtask": "release pen"}, {"step": 8, "subtask": "grasp book"}, {"step": 9, "subtask": "move grasped book out"}, {"step": 10, "subtask": "move grasped book to next to pen"}, {"step": 11, "subtask": "release book"}, {"step": 12, "subtask": "grasp key"}, {"step": 13, "subtask": "move grasped key out"}, {"step": 14, "subtask": "move grasped key to top of book"}, {"step": 15, "subtask": "release key"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a46a657d90cfb88bc5dbc5ae.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0177/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train/5", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 177, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_02_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 514, "role": "error_localization"}}
{"state": {"task": "put the item that is between the sink and stove into the cupboard", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon into the cupboard"}, {"step": 6, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0178/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 178, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 334, "role": "error_localization"}}
{"state": {"task": "Put the item closest to the sink into the sink", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to sink"}, {"step": 3, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0180/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 180, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 278, "role": "error_localization"}}
{"state": {"task": "take spatula off plate sink, then place it on the counter, then pick it up again, then return it to the plate sink", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula out"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "release spatula"}, {"step": 6, "subtask": "move grasped spatula to counter"}, {"step": 7, "subtask": "release spatula"}, {"step": 8, "subtask": "grasp spatula"}, {"step": 9, "subtask": "move grasped spatula to plate sink"}, {"step": 10, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/09c16a5ebc339db4921d2890.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0183/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spatula_off_plate_sink_train/21", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 183, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spatula_off_plate_sink_train", "episode_id": "21", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 411, "role": "error_localization"}}
{"state": {"task": "cook the red pepper in the pot with the oven for 5 minutes, put the pot on the stove, close the oven, then remove the red pepper from the pot, put it on the table, put the pot in the sink", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to pot"}, {"step": 6, "subtask": "release red pepper"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to oven"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp oven door handle"}, {"step": 11, "subtask": "move grasped door handle to close position"}, {"step": 12, "subtask": "release door handle"}, {"step": 13, "subtask": "wait 5 minutes"}, {"step": 14, "subtask": "grasp oven door handle"}, {"step": 15, "subtask": "move grasped door handle to open position"}, {"step": 16, "subtask": "release door handle"}, {"step": 17, "subtask": "grasp pot"}, {"step": 18, "subtask": "move grasped pot to stove"}, {"step": 19, "subtask": "release pot"}, {"step": 20, "subtask": "grasp oven door handle"}, {"step": 21, "subtask": "move grasped door handle to close position"}, {"step": 22, "subtask": "release door handle"}, {"step": 23, "subtask": "grasp red pepper"}, {"step": 24, "subtask": "move grasped red pepper to table"}, {"step": 25, "subtask": "release red pepper"}, {"step": 26, "subtask": "grasp spatula"}, {"step": 27, "subtask": "move grasped spatula to sink"}, {"step": 28, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0184/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 184, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 752, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the can that is the closest to the cyan towel", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cyan towel"}, {"step": 11, "subtask": "move grasped cyan towel to upper right corner"}, {"step": 12, "subtask": "release cyan towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0187/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 187, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 466, "role": "error_localization"}}
{"state": {"task": "place the wok on the purple napkin, then move it to the stove, then back to the napkin, then next to the napkin.", "plan": [{"step": 1, "subtask": "grasp wok"}, {"step": 2, "subtask": "move grasped wok to purple napkin"}, {"step": 3, "subtask": "release wok"}, {"step": 4, "subtask": "grasp wok"}, {"step": 5, "subtask": "move grasped wok to stove"}, {"step": 6, "subtask": "grasp wok"}, {"step": 7, "subtask": "grasp wok"}, {"step": 8, "subtask": "move grasped wok to purple napkin"}, {"step": 9, "subtask": "release wok"}, {"step": 10, "subtask": "release wok"}, {"step": 11, "subtask": "move grasped wok next to purple napkin"}, {"step": 12, "subtask": "release wok"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7f91aada8a55f5a3afb809ef.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0188/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_101_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 188, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_101_train", "episode_id": "17", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "Remove blueberries from plate, place them on counter, then return them to plate, then move them to sink again", "plan": [{"step": 1, "subtask": "release blueberries"}, {"step": 2, "subtask": "move grasped blueberries to counter"}, {"step": 3, "subtask": "release blueberries"}, {"step": 4, "subtask": "grasp blueberries"}, {"step": 5, "subtask": "move grasped blueberries to plate"}, {"step": 6, "subtask": "grasp blueberries"}, {"step": 7, "subtask": "grasp blueberries"}, {"step": 8, "subtask": "move grasped blueberries to sink"}, {"step": 9, "subtask": "release blueberries"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/982439ed1a963a94e8513cfe.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0193/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 193, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "1", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 391, "role": "error_localization"}}
{"state": {"task": "Move the green object into the silver pot, then out of the pot, then place it beside the pot, then back into the pot.", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object to silver pot"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object out of silver pot"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "grasp green object"}, {"step": 8, "subtask": "move grasped green object on top of silver pot"}, {"step": 9, "subtask": "release green object"}, {"step": 10, "subtask": "grasp green object"}, {"step": 11, "subtask": "move grasped green object to silver pot"}, {"step": 12, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0194/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 194, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "15", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 455, "role": "error_localization"}}
{"state": {"task": "open the oven, then move the pot to the open oven door, then close the oven, then move the pan to the open oven door", "plan": [{"step": 1, "subtask": "grasp oven handle"}, {"step": 2, "subtask": "move grasped oven handle to oven's open position"}, {"step": 3, "subtask": "release oven handle"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "move grasped pot to open oven door"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp oven handle"}, {"step": 8, "subtask": "move grasped oven handle to oven's closed position"}, {"step": 9, "subtask": "release oven handle"}, {"step": 10, "subtask": "move grasped pan to open oven door"}, {"step": 11, "subtask": "release pan"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3b82692e6e4e5950f12ad350.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0195/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_close_oven_train/38", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 195, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_close_oven_train", "episode_id": "38", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "cook the red pepper in the pot with the oven for 5 minutes, put the pot on the stove, close the oven, then remove the red pepper from the pot, put it on the table, put the pot in the sink", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "move grasped red apple to pot"}, {"step": 6, "subtask": "release red apple"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to oven"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp oven door handle"}, {"step": 11, "subtask": "move grasped door handle to close position"}, {"step": 12, "subtask": "release door handle"}, {"step": 13, "subtask": "wait 5 minutes"}, {"step": 14, "subtask": "grasp oven door handle"}, {"step": 15, "subtask": "move grasped door handle to open position"}, {"step": 16, "subtask": "release door handle"}, {"step": 17, "subtask": "grasp pot"}, {"step": 18, "subtask": "move grasped pot to stove"}, {"step": 19, "subtask": "release pot"}, {"step": 20, "subtask": "grasp oven door handle"}, {"step": 21, "subtask": "move grasped door handle to close position"}, {"step": 22, "subtask": "release door handle"}, {"step": 23, "subtask": "grasp red pepper"}, {"step": 24, "subtask": "move grasped red pepper to table"}, {"step": 25, "subtask": "release red pepper"}, {"step": 26, "subtask": "grasp pot"}, {"step": 27, "subtask": "move grasped pot to sink"}, {"step": 28, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0196/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 196, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 749, "role": "error_localization"}}
{"state": {"task": "put cucumber in cup, then take it out, then put it back in", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber to cup"}, {"step": 3, "subtask": "move grasped cucumber to cup"}, {"step": 4, "subtask": "grasp cucumber"}, {"step": 5, "subtask": "move grasped cucumber out"}, {"step": 6, "subtask": "release cucumber"}, {"step": 7, "subtask": "grasp cucumber"}, {"step": 8, "subtask": "release cucumber"}, {"step": 9, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/dbce06c934d3d93ffac65ae1.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0199/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/34", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 199, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "34", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 373, "role": "error_localization"}}
{"state": {"task": "Move the green toy to the bottom of the table, then to the top, then to the left side, then back to the bottom.", "plan": [{"step": 1, "subtask": "grasp green toy"}, {"step": 2, "subtask": "move grasped green toy to the bottom of the table"}, {"step": 3, "subtask": "release green toy"}, {"step": 4, "subtask": "grasp green toy"}, {"step": 5, "subtask": "move grasped green toy to the top of the table"}, {"step": 6, "subtask": "release green toy"}, {"step": 7, "subtask": "grasp green toy"}, {"step": 8, "subtask": "release green toy"}, {"step": 9, "subtask": "move grasped green toy to the bottom of the table"}, {"step": 10, "subtask": "release green toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b98ea03009591c137394fc15.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0200/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 200, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_21_train", "episode_id": "4", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 423, "role": "error_localization"}}
{"state": {"task": "put the item that is between the sink and stove into the cupboard", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper into the cupboard"}, {"step": 6, "subtask": "release red pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0194e253fb4e7c60e6d74753.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0202/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/22", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 202, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "22", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 334, "role": "error_localization"}}
{"state": {"task": "move the cloth to right from left, then fold it, then unfold it", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the right"}, {"step": 3, "subtask": "bring it back to its original position"}, {"step": 4, "subtask": "release cloth"}, {"step": 5, "subtask": "grasp cloth"}, {"step": 6, "subtask": "fold cloth"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "unfold cloth"}, {"step": 10, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/92099ed957f0d8169ef126c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0203/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 203, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_11_train", "episode_id": "15", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "move the orange cloth to the left burner", "plan": [{"step": 1, "subtask": "move grasped orange cloth to the left burner"}, {"step": 2, "subtask": "pick up orange cloth"}, {"step": 3, "subtask": "release orange cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7bbd5336fef654e1c6796542.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0204/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_02_train/3", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 204, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_02_train", "episode_id": "3", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 278, "role": "error_localization"}}
{"state": {"task": "heat the bread for 25 minutes in the oven, then put it back in the orange bowl", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp bread"}, {"step": 5, "subtask": "move grasped bread to oven"}, {"step": 6, "subtask": "release bread"}, {"step": 7, "subtask": "grasp oven door handle"}, {"step": 8, "subtask": "move grasped door handle to close position"}, {"step": 9, "subtask": "release door handle"}, {"step": 10, "subtask": "wait 25 minutes"}, {"step": 11, "subtask": "grasp bowl handle"}, {"step": 12, "subtask": "move grasped door handle to open position"}, {"step": 13, "subtask": "release door handle"}, {"step": 14, "subtask": "grasp bread"}, {"step": 15, "subtask": "move grasped bread to orange bowl"}, {"step": 16, "subtask": "release bread"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0205/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 205, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 512, "role": "error_localization"}}
{"state": {"task": "remove the chocolate bar from the yellow cloth, then place it on the table, then return it to the cloth, then remove it again", "plan": [{"step": 1, "subtask": "grasp chocolate bar"}, {"step": 2, "subtask": "move grasped chocolate bar out of yellow cloth"}, {"step": 3, "subtask": "release chocolate bar"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth to table"}, {"step": 6, "subtask": "release yellow cloth"}, {"step": 7, "subtask": "grasp chocolate bar"}, {"step": 8, "subtask": "move grasped chocolate bar to yellow cloth"}, {"step": 9, "subtask": "release chocolate bar"}, {"step": 10, "subtask": "grasp chocolate bar"}, {"step": 11, "subtask": "move grasped chocolate bar out of yellow cloth"}, {"step": 12, "subtask": "release chocolate bar"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4afdf0776028e51c3aec77d2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0206/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_76_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 206, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_76_train", "episode_id": "11", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 453, "role": "error_localization"}}
{"state": {"task": "move the blue spatula on the left burner, then move it to the right burner, then move it back to the left burner", "plan": [{"step": 1, "subtask": "grasp blue spatula"}, {"step": 2, "subtask": "move grasped blue spatula to middle burner"}, {"step": 3, "subtask": "release blue spatula"}, {"step": 4, "subtask": "grasp blue spatula"}, {"step": 5, "subtask": "move grasped blue spatula to right burner"}, {"step": 6, "subtask": "release blue spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/783b3ec5d8c46e163ee8e7c4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0207/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 207, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "move the cloth from the bottom right to the top left of the table, then fold it in half, then move it back to the bottom right", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the top left of the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold grasped cloth in half"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "move grasped cloth to the bottom right of the table"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bf568029e1d1a3734b86d995.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0208/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 208, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train", "episode_id": "17", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 398, "role": "error_localization"}}
{"state": {"task": "take spoon out of bowl, then place it on the dish rack, then move it to the counter", "plan": [{"step": 1, "subtask": "move grasped spoon to dish rack"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "grasp spoon"}, {"step": 4, "subtask": "release spoon"}, {"step": 5, "subtask": "grasp spoon"}, {"step": 6, "subtask": "move grasped spoon to counter"}, {"step": 7, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8910d4c69f20807a610ef6e3.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0209/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train/18", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 209, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_spoon_out_of_bowl_sink_train", "episode_id": "18", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "move the cloth to the bottom right, then to the top left, then fold it, then to the center", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to the bottom right"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "release cloth"}, {"step": 6, "subtask": "move grasped cloth to the top left"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "rotate grasped cloth"}, {"step": 10, "subtask": "push down"}, {"step": 11, "subtask": "release cloth"}, {"step": 12, "subtask": "grasp cloth"}, {"step": 13, "subtask": "move grasped cloth to the center"}, {"step": 14, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/67c7b1f2958e652cec2c5c34.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0212/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 212, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_27_train", "episode_id": "17", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 464, "role": "error_localization"}}
{"state": {"task": "put the cans in the cupboard, put the bread on the table, and put the red towel, and the orange bowl in the sink", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to cupboard"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to cupboard"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped bread to table"}, {"step": 12, "subtask": "release bread"}, {"step": 13, "subtask": "grasp red towel"}, {"step": 14, "subtask": "move grasped red towel to sink"}, {"step": 15, "subtask": "release red towel"}, {"step": 16, "subtask": "grasp orange bowl"}, {"step": 17, "subtask": "move grasped orange bowl to sink"}, {"step": 18, "subtask": "release orange bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0213/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 213, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 550, "role": "error_localization"}}
{"state": {"task": "take the silver pot lid, place it on the silver pot, then remove it, then place it again", "plan": [{"step": 1, "subtask": "grasp silver pot lid"}, {"step": 2, "subtask": "move grasped silver pot lid to silver pot"}, {"step": 3, "subtask": "release silver pot lid"}, {"step": 4, "subtask": "move grasped silver pot lid to silver pot"}, {"step": 5, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8f732e0d674b04d9d22289f8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0214/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train/0", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 214, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_04_train", "episode_id": "0", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 330, "role": "error_localization"}}
{"state": {"task": "place the yellow brush on the blue cloth, then move the blue cloth to the green tray, then place the yellow brush back on the blue cloth.", "plan": [{"step": 1, "subtask": "grasp yellow brush"}, {"step": 2, "subtask": "move grasped yellow brush to blue cloth"}, {"step": 3, "subtask": "release yellow brush"}, {"step": 4, "subtask": "grasp blue cloth"}, {"step": 5, "subtask": "release blue cloth"}, {"step": 6, "subtask": "grasp yellow brush"}, {"step": 7, "subtask": "move grasped yellow brush to blue cloth"}, {"step": 8, "subtask": "release yellow brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0ae8aa96ce4023993413e5e8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0215/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_93_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 215, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_93_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 380, "role": "error_localization"}}
{"state": {"task": "move the lemon on the right side of the stove, then to the left side, then back to the right side, then to the front side of the stove", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to the right side of the stove"}, {"step": 3, "subtask": "release lemon"}, {"step": 4, "subtask": "grasp lemon"}, {"step": 5, "subtask": "move grasped lemon to the right side of the stove"}, {"step": 6, "subtask": "release lemon"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped lemon to the front side of the stove"}, {"step": 9, "subtask": "release lemon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6298e643302c41acc33c2f7d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0217/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_24_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 217, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_24_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "put the yellow towel at the bottom of the left burner", "plan": [{"step": 1, "subtask": "pick up yellow towel"}, {"step": 2, "subtask": "move grasped yellow towel to the bottom of the left burner"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release yellow towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3ade824fdf797412725045a9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0218/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/12", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 218, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "12", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 303, "role": "error_localization"}}
{"state": {"task": "Put cup to drying rack from sink, then move plate from sink to drying rack, then fork from sink to drying rack", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "release cup"}, {"step": 3, "subtask": "grasp plate"}, {"step": 4, "subtask": "move grasped plate to drying rack"}, {"step": 5, "subtask": "release plate"}, {"step": 6, "subtask": "move grasped fork to drying rack"}, {"step": 7, "subtask": "release fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f0350fc1da496a4bb9b0fc6d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0219/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/5", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 219, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "5", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "place the eggplant on top of the green cloth, then move the cloth with the eggplant to the table, then remove the eggplant and place it on the floor", "plan": [{"step": 1, "subtask": "grasp eggplant"}, {"step": 2, "subtask": "move grasped eggplant to top of the green cloth"}, {"step": 3, "subtask": "release eggplant"}, {"step": 4, "subtask": "grasp green cloth"}, {"step": 5, "subtask": "move grasped green cloth to the floor"}, {"step": 6, "subtask": "release green cloth"}, {"step": 7, "subtask": "grasp eggplant"}, {"step": 8, "subtask": "move grasped eggplant to the table"}, {"step": 9, "subtask": "release eggplant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2598fd4c6c2589d34bf8a1d9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0222/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_39_train/5", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 222, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_39_train", "episode_id": "5", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 408, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tap handle"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp left burner"}, {"step": 10, "subtask": "move grasped object to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp left burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner knob"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0224/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 224, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 610, "role": "error_localization"}}
{"state": {"task": "Put spatula on plate, then on sink, then back on plate, then remove another object from drawer and place it on the sink.", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula to plate"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to sink"}, {"step": 6, "subtask": "release spatula"}, {"step": 7, "subtask": "grasp spatula"}, {"step": 8, "subtask": "move grasped spatula to plate"}, {"step": 9, "subtask": "move it back to the original position"}, {"step": 10, "subtask": "release spatula"}, {"step": 11, "subtask": "grasp drawer handle"}, {"step": 12, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 13, "subtask": "release drawer handle"}, {"step": 14, "subtask": "grasp object"}, {"step": 15, "subtask": "move grasped object to sink"}, {"step": 16, "subtask": "release object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9f3ffe0b1b248aeee46759a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0225/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 225, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "2", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 521, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the lower side of the right stove, then move it to the upper side, then back to the lower side, and finally to the right side of the stove.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the lower side of the right stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the upper side of the right stove"}, {"step": 6, "subtask": "move it back to the original position"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to the lower side of the right stove"}, {"step": 10, "subtask": "release silver pot"}, {"step": 11, "subtask": "grasp silver pot"}, {"step": 12, "subtask": "move grasped silver pot to the right side of the stove"}, {"step": 13, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b5fb632f98ec8ac70cfe3516.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0226/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_03_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 226, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_03_train", "episode_id": "17", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 501, "role": "error_localization"}}
{"state": {"task": "move the blue spoon to the top right corner of the stove, then to the top left corner, then to the bottom right corner, then back to the top right corner", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the top right corner of the stove"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the top left corner of the stove"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "release blue spoon"}, {"step": 9, "subtask": "move grasped blue spoon to the bottom right corner of the stove"}, {"step": 10, "subtask": "release blue spoon"}, {"step": 11, "subtask": "grasp blue spoon"}, {"step": 12, "subtask": "move grasped blue spoon to the top right corner of the stove"}, {"step": 13, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8dc55a1db643dd8960bc9dcc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0227/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_95_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 227, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_95_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 495, "role": "error_localization"}}
{"state": {"task": "Take the potato, move it to the right side of the stove, then to the left, then back to the right.", "plan": [{"step": 1, "subtask": "grasp potato"}, {"step": 2, "subtask": "move grasped potato to the right side of the stove"}, {"step": 3, "subtask": "release potato"}, {"step": 4, "subtask": "grasp stove"}, {"step": 5, "subtask": "move grasped potato to the left side of the stove"}, {"step": 6, "subtask": "release potato"}, {"step": 7, "subtask": "grasp stove"}, {"step": 8, "subtask": "move grasped potato to the right side of the stove"}, {"step": 9, "subtask": "release potato"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c00075ccf1d64eba9612b3cf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0228/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 228, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "10", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 398, "role": "error_localization"}}
{"state": {"task": "Remove the spoon from the orange pot, then place it on the table, pick it up again, and return it to the orange pot.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of orange pot"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to orange pot"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to table"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4b15d6af50f2480076658224.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0229/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 229, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 389, "role": "error_localization"}}
{"state": {"task": "Put the yellow towel on the left burner, then move it to the right burner, then fold it neatly on the counter.", "plan": [{"step": 1, "subtask": "grasp yellow towel"}, {"step": 2, "subtask": "move grasped yellow towel to the right burner"}, {"step": 3, "subtask": "release yellow towel"}, {"step": 4, "subtask": "grasp yellow towel"}, {"step": 5, "subtask": "move grasped yellow towel to the left burner"}, {"step": 6, "subtask": "release yellow towel"}, {"step": 7, "subtask": "grasp yellow towel"}, {"step": 8, "subtask": "move grasped yellow towel to the counter"}, {"step": 9, "subtask": "fold grasped yellow towel"}, {"step": 10, "subtask": "release yellow towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/70c5fe5722bd30d78ae38041.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0232/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/0", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 232, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "0", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 415, "role": "error_localization"}}
{"state": {"task": "move the cloth to top left, then fold it, then move it to bottom right, then unfold it", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to bottom left"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "fold grasped cloth"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to bottom right"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "unfold grasped cloth"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/10cbb8db026437d74d79a7b6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0233/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train/12", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 233, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_32_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 428, "role": "error_localization"}}
{"state": {"task": "put the cloth on the table to the burner", "plan": [{"step": 1, "subtask": "pick up cloth"}, {"step": 2, "subtask": "move grasped cloth to the table"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "pick up cloth"}, {"step": 5, "subtask": "move grasped cloth to the burner"}, {"step": 6, "subtask": "move it back to the original position"}, {"step": 7, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6289f95cd3b7f3e99705c663.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0234/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train/5", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 234, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_26_train", "episode_id": "5", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "Move the blue scoop to the top left side of the stove, then to the top right side, then to the bottom left side, then to the bottom right side.", "plan": [{"step": 1, "subtask": "grasp blue scoop"}, {"step": 2, "subtask": "move grasped blue scoop to the top left side of the stove"}, {"step": 3, "subtask": "release blue scoop"}, {"step": 4, "subtask": "grasp blue scoop"}, {"step": 5, "subtask": "move grasped blue scoop to the top right side of the stove"}, {"step": 6, "subtask": "release blue scoop"}, {"step": 7, "subtask": "grasp blue scoop"}, {"step": 8, "subtask": "release blue scoop"}, {"step": 9, "subtask": "move grasped blue scoop to the bottom left side of the stove"}, {"step": 10, "subtask": "release blue scoop"}, {"step": 11, "subtask": "grasp blue scoop"}, {"step": 12, "subtask": "move grasped blue scoop to the bottom right side of the stove"}, {"step": 13, "subtask": "release blue scoop"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4330bced814cab052c9319be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0235/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_27_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 235, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_27_train", "episode_id": "10", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 494, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally back to the top right corner.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "move grasped cloth to top right corner"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "grasp cloth"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0238/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 238, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 453, "role": "error_localization"}}
{"state": {"task": "put the cans in the left cupboard, put the bread on the table, and put the red towel, and the orange bowl in the sink", "plan": [{"step": 1, "subtask": "grasp bottom cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to left cupboard"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp another can"}, {"step": 8, "subtask": "move grasped can to left cupboard"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped bread to table"}, {"step": 12, "subtask": "release bread"}, {"step": 13, "subtask": "grasp red towel"}, {"step": 14, "subtask": "move grasped red towel to sink"}, {"step": 15, "subtask": "release red towel"}, {"step": 16, "subtask": "grasp orange bowl"}, {"step": 17, "subtask": "move grasped orange bowl to sink"}, {"step": 18, "subtask": "release orange bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0239/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 239, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 552, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up bowl"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up corn"}, {"step": 5, "subtask": "release corn"}, {"step": 6, "subtask": "move grasped corn to sink"}, {"step": 7, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4564bea6b84939cd0266e47d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0242/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 242, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 331, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp left burner knob"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp bowl"}, {"step": 10, "subtask": "move grasped bowl to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp left burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0244/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 244, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 609, "role": "error_localization"}}
{"state": {"task": "move the red bell pepper to the right side of burner, then to the front left, then to the back right, then back to the right side", "plan": [{"step": 1, "subtask": "grasp red bell pepper"}, {"step": 2, "subtask": "move grasped red bell pepper to the right side of burner"}, {"step": 3, "subtask": "release red bell pepper"}, {"step": 4, "subtask": "grasp red bell pepper"}, {"step": 5, "subtask": "move grasped red bell pepper to the front left"}, {"step": 6, "subtask": "release red bell pepper"}, {"step": 7, "subtask": "grasp red bell pepper"}, {"step": 8, "subtask": "move grasped red bell pepper to the back right"}, {"step": 9, "subtask": "release red bell pepper"}, {"step": 10, "subtask": "grasp red bell pepper"}, {"step": 11, "subtask": "move grasped red bell pepper to the right side of burner"}, {"step": 12, "subtask": "return it to its original position"}, {"step": 13, "subtask": "release red bell pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2f70c49451ef8db101e323c6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0248/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 248, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_29_train", "episode_id": "11", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 494, "role": "error_localization"}}
{"state": {"task": "move the croissant to the upper right side of the stove, then to the lower right side, then to the upper left side, then back to the upper right side", "plan": [{"step": 1, "subtask": "grasp croissant"}, {"step": 2, "subtask": "move grasped croissant to the upper right side of the stove"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release croissant"}, {"step": 5, "subtask": "grasp croissant"}, {"step": 6, "subtask": "move grasped croissant to the lower right side of the stove"}, {"step": 7, "subtask": "release croissant"}, {"step": 8, "subtask": "grasp croissant"}, {"step": 9, "subtask": "move grasped croissant to the upper left side of the stove"}, {"step": 10, "subtask": "release croissant"}, {"step": 11, "subtask": "grasp croissant"}, {"step": 12, "subtask": "move grasped croissant to the upper right side of the stove"}, {"step": 13, "subtask": "release croissant"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2fa3823f2d16818c90494012.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0250/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 250, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train", "episode_id": "14", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 499, "role": "error_localization"}}
{"state": {"task": "unfold the cloth from top right to bottom left", "plan": [{"step": 1, "subtask": "move grasped cloth from top right to bottom left"}, {"step": 2, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9441a43315c831de0b097470.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0254/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_06_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 254, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_06_train", "episode_id": "14", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 263, "role": "error_localization"}}
{"state": {"task": "Remove the silver lid from the silver pot, then place it on the table, and put the silver pot back under the sink.", "plan": [{"step": 1, "subtask": "grasp silver lid"}, {"step": 2, "subtask": "move grasped silver lid out of silver pot"}, {"step": 3, "subtask": "move grasped silver pot to under sink"}, {"step": 4, "subtask": "grasp silver lid"}, {"step": 5, "subtask": "move grasped silver lid to table"}, {"step": 6, "subtask": "release silver lid"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "release silver lid"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/be32103799c792ede7b439c8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0256/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train/0", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 256, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_00_train", "episode_id": "0", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 396, "role": "error_localization"}}
{"state": {"task": "put the silver pot lid on the top of the silver pot, then remove it, then place it on the side of the silver pot", "plan": [{"step": 1, "subtask": "grasp silver pot lid"}, {"step": 2, "subtask": "release silver pot lid"}, {"step": 3, "subtask": "move grasped silver pot lid to the side of the silver pot"}, {"step": 4, "subtask": "release silver pot lid"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/58e7d8debd78055fa8271394.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0257/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 257, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_36_train", "episode_id": "14", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 319, "role": "error_localization"}}
{"state": {"task": "remove the wooden spatula to the basket and place it from the side of the right burner", "plan": [{"step": 1, "subtask": "pick up wooden spatula"}, {"step": 2, "subtask": "release wooden spatula"}, {"step": 3, "subtask": "move grasped wooden spatula to the basket"}, {"step": 4, "subtask": "release wooden spatula"}, {"step": 5, "subtask": "move grasped wooden spatula from the side of the right burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b369b2e6e20494b0b44c6704.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0263/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_03_train/0", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 263, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_03_train", "episode_id": "0", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 331, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put the yellow can to the upper right corner", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp tomato can"}, {"step": 11, "subtask": "move grasped can to upper right corner"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0265/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 265, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 457, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the upper middle of the table, then to the lower left corner, then back to the center", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the upper middle of the table"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "grasp silver pot"}, {"step": 7, "subtask": "move grasped silver pot to the center of the table"}, {"step": 8, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8d6dfedb78b3811df610b491.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0268/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_12_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 268, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_12_train", "episode_id": "14", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 382, "role": "error_localization"}}
{"state": {"task": "Move the yellow object to the bottom right edge of burners, then to the top left edge, then back to the bottom right edge.", "plan": [{"step": 1, "subtask": "grasp yellow object"}, {"step": 2, "subtask": "move grasped yellow object to bottom right edge of burners"}, {"step": 3, "subtask": "move it back to the original position"}, {"step": 4, "subtask": "release yellow object"}, {"step": 5, "subtask": "grasp yellow object"}, {"step": 6, "subtask": "move grasped yellow object to top left edge of burners"}, {"step": 7, "subtask": "release yellow object"}, {"step": 8, "subtask": "grasp yellow object"}, {"step": 9, "subtask": "move grasped yellow object to bottom right edge of burners"}, {"step": 10, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5256d032f07b2e7b5bb8362c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0269/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train/19", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 269, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train", "episode_id": "19", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 430, "role": "error_localization"}}
{"state": {"task": "move the blue spatula to the left side of the burner, then to the right side, then to the center, then back to the left side", "plan": [{"step": 1, "subtask": "grasp blue spatula"}, {"step": 2, "subtask": "move grasped blue spatula to the left side of the burner"}, {"step": 3, "subtask": "release blue spatula"}, {"step": 4, "subtask": "grasp blue spatula"}, {"step": 5, "subtask": "move grasped blue spatula to the right side of the burner"}, {"step": 6, "subtask": "release blue spatula"}, {"step": 7, "subtask": "grasp blue spatula"}, {"step": 8, "subtask": "move grasped blue spatula to the right side of the burner"}, {"step": 9, "subtask": "release blue spatula"}, {"step": 10, "subtask": "grasp blue spatula"}, {"step": 11, "subtask": "move grasped blue spatula to the left side of the burner"}, {"step": 12, "subtask": "release blue spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/14f4a698e896767d3f97c4b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0273/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_65_train/12", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 273, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_65_train", "episode_id": "12", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 483, "role": "error_localization"}}
{"state": {"task": "Move the red cloth to the lower left corner of the stove, then to the upper left corner, then to the upper right corner, then back to the lower left corner.", "plan": [{"step": 1, "subtask": "grasp red cloth"}, {"step": 2, "subtask": "move grasped red cloth to the upper left corner of the stove"}, {"step": 3, "subtask": "release red cloth"}, {"step": 4, "subtask": "grasp red cloth"}, {"step": 5, "subtask": "move grasped red cloth to the lower left corner of the stove"}, {"step": 6, "subtask": "release red cloth"}, {"step": 7, "subtask": "grasp red cloth"}, {"step": 8, "subtask": "move grasped red cloth to the upper right corner of the stove"}, {"step": 9, "subtask": "release red cloth"}, {"step": 10, "subtask": "grasp red cloth"}, {"step": 11, "subtask": "move grasped red cloth to the lower left corner of the stove"}, {"step": 12, "subtask": "release red cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0277/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 277, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 479, "role": "error_localization"}}
{"state": {"task": "Move the orange toy to the left side of the table, then to the center, then back to the left side.", "plan": [{"step": 1, "subtask": "grasp orange toy"}, {"step": 2, "subtask": "grasp orange toy"}, {"step": 3, "subtask": "release orange toy"}, {"step": 4, "subtask": "move grasped orange toy to the left side of the table"}, {"step": 5, "subtask": "move grasped orange toy to the center of the table"}, {"step": 6, "subtask": "release orange toy"}, {"step": 7, "subtask": "grasp orange toy"}, {"step": 8, "subtask": "move grasped orange toy to the left side of the table"}, {"step": 9, "subtask": "release orange toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/205d1ee7a28a1e9f41bd71f7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0279/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 279, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 405, "role": "error_localization"}}
{"state": {"task": "put strawberry in bowl, then take bowl out of sink onto the table, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped object to bowl"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "move grasped bowl to the table"}, {"step": 7, "subtask": "release bowl"}, {"step": 8, "subtask": "grasp cabinet door handle"}, {"step": 9, "subtask": "move grasped door handle to open position"}, {"step": 10, "subtask": "release door handle"}, {"step": 11, "subtask": "grasp spoon"}, {"step": 12, "subtask": "move grasped spoon to cabinet"}, {"step": 13, "subtask": "release spoon"}, {"step": 14, "subtask": "grasp fork"}, {"step": 15, "subtask": "move grasped object to bowl"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0281/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 281, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 510, "role": "error_localization"}}
{"state": {"task": "Move the silver pot on the top of the orange towel, then place it next to the silver fork, then back on the orange towel.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the top of the orange towel"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top of the orange towel"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the top of the silver fork"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bb027c35653a12daa376ee60.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0283/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_21_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 283, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_21_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 410, "role": "error_localization"}}
{"state": {"task": "move the can to the center of the table, then to the front left, back right, and front right corners.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the center of the table"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the back left corner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the back right corner"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "move grasped can to the front right corner"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/577bf5296cef880c2fd84381.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0284/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train/16", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 284, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_19_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "Put the cup to the counter from the sink, then to the stove, then to the fridge handle, then back to the counter.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to fridge handle"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to counter"}, {"step": 12, "subtask": "move it back to the original position"}, {"step": 13, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b321bf8039405b2c9c4ac518.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0285/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/67", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 285, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "67", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 456, "role": "error_localization"}}
{"state": {"task": "Place the blue spoon on top of the orange cloth, then move the spoon to the center of the table, then back to the cloth.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to center of the table"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to top of the orange cloth"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to center of the table"}, {"step": 9, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/90bf73d2486a48559acb5f8d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0287/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_67_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 287, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_67_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 405, "role": "error_localization"}}
{"state": {"task": "take the yellow bottle with black label and put it on the right burner, then to the left burner, then back to the right burner", "plan": [{"step": 1, "subtask": "grasp yellow bottle with black label"}, {"step": 2, "subtask": "move grasped yellow bottle with black label to the left burner"}, {"step": 3, "subtask": "release yellow bottle with black label"}, {"step": 4, "subtask": "grasp yellow bottle with black label"}, {"step": 5, "subtask": "move grasped yellow bottle with black label to the left burner"}, {"step": 6, "subtask": "release yellow bottle with black label"}, {"step": 7, "subtask": "grasp yellow bottle with black label"}, {"step": 8, "subtask": "move grasped yellow bottle with black label to the left burner"}, {"step": 9, "subtask": "release yellow bottle with black label"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cf8e8869f198215689760a2f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0294/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train/3", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 294, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_07_train", "episode_id": "3", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 428, "role": "error_localization"}}
{"state": {"task": "Move the yellow cloth from the middle front of the burners to behind the right side of the burners, then to the left side, then back to the middle front.", "plan": [{"step": 1, "subtask": "grasp yellow cloth"}, {"step": 2, "subtask": "move grasped yellow cloth to behind the left side of the burners"}, {"step": 3, "subtask": "release yellow cloth"}, {"step": 4, "subtask": "grasp yellow cloth"}, {"step": 5, "subtask": "move grasped yellow cloth to the left side of the burners"}, {"step": 6, "subtask": "release yellow cloth"}, {"step": 7, "subtask": "grasp yellow cloth"}, {"step": 8, "subtask": "move grasped yellow cloth to the middle front of the burners"}, {"step": 9, "subtask": "release yellow cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b13a7904f6d38c43dac100f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0297/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 297, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_14_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 421, "role": "error_localization"}}
{"state": {"task": "Put the blue spoon on the top of the green towel, then move it to the left edge, then to the right edge, then back to the center.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the top of the green towel"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the left edge of the green towel"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to the left edge of the green towel"}, {"step": 9, "subtask": "release blue spoon"}, {"step": 10, "subtask": "grasp blue spoon"}, {"step": 11, "subtask": "move grasped blue spoon to the center of the green towel"}, {"step": 12, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c728ac183e0042e2196d76b8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0299/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_31_train/6", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 299, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_31_train", "episode_id": "6", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 474, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl, then take bowl out of sink onto the table, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "move grasped bowl to the table"}, {"step": 7, "subtask": "release bowl"}, {"step": 8, "subtask": "grasp oven door handle"}, {"step": 9, "subtask": "move grasped door handle to open position"}, {"step": 10, "subtask": "release door handle"}, {"step": 11, "subtask": "grasp spoon"}, {"step": 12, "subtask": "move grasped spoon to cabinet"}, {"step": 13, "subtask": "release spoon"}, {"step": 14, "subtask": "grasp fork"}, {"step": 15, "subtask": "move grasped object to bowl"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0301/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 301, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 511, "role": "error_localization"}}
{"state": {"task": "move the green object into the silver pot, then remove a lid from a nearby drawer and place it on top of the pot, then take the green object out of the pot", "plan": [{"step": 1, "subtask": "move grasped lid to top of silver pot"}, {"step": 2, "subtask": "move grasped green object to silver pot"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp drawer handle"}, {"step": 5, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 6, "subtask": "release drawer handle"}, {"step": 7, "subtask": "grasp lid"}, {"step": 8, "subtask": "grasp green object"}, {"step": 9, "subtask": "release lid"}, {"step": 10, "subtask": "grasp green object"}, {"step": 11, "subtask": "move grasped green object out of silver pot"}, {"step": 12, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23f30f0833d19ca2a5d85992.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0303/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_52_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 303, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_52_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 462, "role": "error_localization"}}
{"state": {"task": "try to pull up the green cloth, then fold it, then put it inside the box", "plan": [{"step": 1, "subtask": "grasp green cloth"}, {"step": 2, "subtask": "move grasped green cloth out"}, {"step": 3, "subtask": "rotate grasped green cloth"}, {"step": 4, "subtask": "move grasped green cloth to floor"}, {"step": 5, "subtask": "release green cloth"}, {"step": 6, "subtask": "grasp green cloth"}, {"step": 7, "subtask": "move grasped green cloth to outside box"}, {"step": 8, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fabe07a6742a375eeff13925.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0307/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/9", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 307, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "close the bottle top", "plan": [{"step": 1, "subtask": "pick up bottle"}, {"step": 2, "subtask": "release bottle"}, {"step": 3, "subtask": "move grasped bottle to upright position"}, {"step": 4, "subtask": "release bottle"}, {"step": 5, "subtask": "pick up bottle top"}, {"step": 6, "subtask": "move grasped bottle top to the bottle's closed position"}, {"step": 7, "subtask": "release bottle top"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f06814aa0250a1830e2561dd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0309/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_20_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 309, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_20_train", "episode_id": "8", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 338, "role": "error_localization"}}
{"state": {"task": "move the silver pot on top of the green cloth, then move it to the wooden tray, then back on top of the green cloth", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "release silver pot"}, {"step": 3, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to wooden tray"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dfd421c8d0c3cbba5e2cc99.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0314/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/3", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 314, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "3", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 404, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from the orange cloth, place it on the table, then return it to the orange cloth, then move it to the green towel.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out of orange cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to table"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to orange cloth"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp orange cloth"}, {"step": 11, "subtask": "move grasped silver pot to the green towel"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0316/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 316, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "remove the pink toy from the top of the drawer", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up pink toy"}, {"step": 3, "subtask": "release pink toy"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/93541c39e97072b2e745847e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0317/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 317, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 273, "role": "error_localization"}}
{"state": {"task": "put cup to counter from sink, then to the stove, then back to the counter, then to the table", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to counter"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "move grasped cup to table"}, {"step": 11, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aba7dfe4d7f485e002ab0ec2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0318/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 318, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 415, "role": "error_localization"}}
{"state": {"task": "remove the blueberries from the plate, then move the plate to the sink, then empty the plate into the sink", "plan": [{"step": 1, "subtask": "move grasped blueberries out"}, {"step": 2, "subtask": "release blueberries"}, {"step": 3, "subtask": "grasp plate"}, {"step": 4, "subtask": "move grasped plate to sink"}, {"step": 5, "subtask": "release plate"}, {"step": 6, "subtask": "push plate forward"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9239e1c25c73603f4f81ddff.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0321/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/0", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 321, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "0", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 336, "role": "error_localization"}}
{"state": {"task": "remove the chicken leg from the silver pot", "plan": [{"step": 1, "subtask": "release chicken leg"}, {"step": 2, "subtask": "move grasped silver pot upright"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "pick up chicken leg"}, {"step": 5, "subtask": "move grasped chicken leg away from silver pot"}, {"step": 6, "subtask": "pick up silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/605c2698db40b23da2e36801.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0322/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 322, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_05_train", "episode_id": "14", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 327, "role": "error_localization"}}
{"state": {"task": "take banana out of colander", "plan": [{"step": 1, "subtask": "move grasped banana away from colander"}, {"step": 2, "subtask": "move grasped colander to a stable surface"}, {"step": 3, "subtask": "release colander"}, {"step": 4, "subtask": "pick up banana"}, {"step": 5, "subtask": "pick up colander"}, {"step": 6, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/71968079ee95315f5cd44abe.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0327/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train/16", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 327, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_dark_wood_take_banana_out_of_colander_train", "episode_id": "16", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 325, "role": "error_localization"}}
{"state": {"task": "move the yellow fish to the silver dish from the left stove", "plan": [{"step": 1, "subtask": "pick up yellow fish"}, {"step": 2, "subtask": "move grasped yellow fish to the silver dish"}, {"step": 3, "subtask": "release yellow fish"}, {"step": 4, "subtask": "release silver dish"}, {"step": 5, "subtask": "move grasped silver dish to the left stove"}, {"step": 6, "subtask": "pick up silver dish"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3c7bf5a9d12d373a77a77f1c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0329/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_27_train/3", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 329, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_27_train", "episode_id": "3", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 334, "role": "error_localization"}}
{"state": {"task": "Remove cup from plate, then place it on the table, then return it to the plate, then move it to the tray.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to table"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to plate"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to tray"}, {"step": 12, "subtask": "return it to its original position"}, {"step": 13, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cf119044eef1f2d1c1f0beaf.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0333/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/39", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 333, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "39", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "Put the item closest to the sink into the sink", "plan": [{"step": 1, "subtask": "grasp potato"}, {"step": 2, "subtask": "move grasped potato to sink"}, {"step": 3, "subtask": "release potato"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3462716262c45c3c4ba73640.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0334/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 334, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_35_train", "episode_id": "1", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 275, "role": "error_localization"}}
{"state": {"task": "close the drawer, open it, take out the pen, put it on the table, close the drawer again", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp drawer handle"}, {"step": 5, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 6, "subtask": "release drawer handle"}, {"step": 7, "subtask": "grasp pen"}, {"step": 8, "subtask": "move grasped pen out"}, {"step": 9, "subtask": "move grasped pen to table"}, {"step": 10, "subtask": "release drawer handle"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d54b05fdb7cf5b7311fd3735.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0335/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train/6", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 335, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_07_train", "episode_id": "6", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 412, "role": "error_localization"}}
{"state": {"task": "move the cloth to top left, then fold it neatly, then unfold it at the center of the table", "plan": [{"step": 1, "subtask": "release cloth"}, {"step": 2, "subtask": "grasp cloth"}, {"step": 3, "subtask": "move grasped cloth to center of table"}, {"step": 4, "subtask": "rotate grasped cloth"}, {"step": 5, "subtask": "push down"}, {"step": 6, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/674a937e2120c2e36734f7eb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0336/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train/37", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 336, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_11_train", "episode_id": "37", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 332, "role": "error_localization"}}
{"state": {"task": "put cup to drying rack from sink", "plan": [{"step": 1, "subtask": "pick up cup"}, {"step": 2, "subtask": "move grasped cup to countertop"}, {"step": 3, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/36d8c7acc6185f164efc472b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0338/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 338, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "4", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 273, "role": "error_localization"}}
{"state": {"task": "Remove the cup from the plate, then place it on the table, then return it to the plate, then take it off again and place it on the plate upside down.", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "release cup"}, {"step": 6, "subtask": "move grasped cup to table"}, {"step": 7, "subtask": "release cup"}, {"step": 8, "subtask": "grasp cup"}, {"step": 9, "subtask": "move grasped cup to plate"}, {"step": 10, "subtask": "release cup"}, {"step": 11, "subtask": "grasp cup"}, {"step": 12, "subtask": "rotate grasped cup"}, {"step": 13, "subtask": "move grasped cup to plate"}, {"step": 14, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/05e174d4bfb7f2c026e589f6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0344/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/30", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 344, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "30", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 476, "role": "error_localization"}}
{"state": {"task": "Remove blueberries from plate, place in bowl, then transfer to sink, then return to plate.", "plan": [{"step": 1, "subtask": "grasp blueberries"}, {"step": 2, "subtask": "move grasped blueberries to bowl"}, {"step": 3, "subtask": "put it back where it was"}, {"step": 4, "subtask": "release blueberries"}, {"step": 5, "subtask": "grasp blueberries"}, {"step": 6, "subtask": "move grasped blueberries to sink"}, {"step": 7, "subtask": "release blueberries"}, {"step": 8, "subtask": "grasp blueberries"}, {"step": 9, "subtask": "move grasped blueberries to plate"}, {"step": 10, "subtask": "release blueberries"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/50178fd78a775935092228cd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0346/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 346, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_blueberries_on_plate_sink_train", "episode_id": "13", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "Move the yellow object to the pot from the left of the left side burner, then return it to its original position, then move it to the right of the right side burner.", "plan": [{"step": 1, "subtask": "grasp yellow object"}, {"step": 2, "subtask": "release yellow object"}, {"step": 3, "subtask": "grasp yellow object"}, {"step": 4, "subtask": "release yellow object"}, {"step": 5, "subtask": "grasp yellow object"}, {"step": 6, "subtask": "move grasped yellow object to the right of the right side burner"}, {"step": 7, "subtask": "release yellow object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8eff56aaacb09f759a791e04.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0347/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 347, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_38_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "Put the spoon in the orange pot, then remove the lid from the pot, then put the spoon back in the pot.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to orange pot"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp pot lid"}, {"step": 5, "subtask": "move grasped pot lid away"}, {"step": 6, "subtask": "release pot lid"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to blue pot"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/aed4303462eb9330157daa5c.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0352/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 352, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_00_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put the yellow can to the upper right corner", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp cloth"}, {"step": 11, "subtask": "move grasped cloth to upper right corner"}, {"step": 12, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0353/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 353, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 456, "role": "error_localization"}}
{"state": {"task": "remove the green object from between the burners, then place it on the table, then return it to the original spot, then place it beside the stove", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object out from between the burners"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object to table"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "grasp green object"}, {"step": 8, "subtask": "move grasped green object to original spot"}, {"step": 9, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6ea37b21854ef0c98071472d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0354/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_24_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 354, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_24_train", "episode_id": "15", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 404, "role": "error_localization"}}
{"state": {"task": "remove the green spatula from the right burner, then place it on the left burner, then return it to the right burner", "plan": [{"step": 1, "subtask": "grasp green spatula"}, {"step": 2, "subtask": "move grasped green spatula out of right burner"}, {"step": 3, "subtask": "release green spatula"}, {"step": 4, "subtask": "grasp green spatula"}, {"step": 5, "subtask": "move grasped green spatula to left burner"}, {"step": 6, "subtask": "release green spatula"}, {"step": 7, "subtask": "grasp green spatula"}, {"step": 8, "subtask": "release green spatula"}, {"step": 9, "subtask": "move grasped green spatula to right burner"}, {"step": 10, "subtask": "release green spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/13acb27e9a90f1e76fe5bdf8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0357/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_51_train/16", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 357, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_51_train", "episode_id": "16", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 423, "role": "error_localization"}}
{"state": {"task": "put into the silver pot the mushroom and the violet cloth, then put the silver pot in the sink", "plan": [{"step": 1, "subtask": "pick up mushroom"}, {"step": 2, "subtask": "move grasped mushroom to silver pot"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to pot"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to sink"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a11e82fa3b7b4b4505aa0993.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0359/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_54_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 359, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_54_train", "episode_id": "13", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 381, "role": "error_localization"}}
{"state": {"task": "move the spoon between the burners, then place it on the stove's back burner, then move it to the front burner", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon between the burners"}, {"step": 3, "subtask": "grasp spoon"}, {"step": 4, "subtask": "release spoon"}, {"step": 5, "subtask": "move grasped spoon to the stove's back burner"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to the front burner"}, {"step": 9, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/624fcc2bfce1768bb84ae25f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0360/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 360, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_38_train", "episode_id": "11", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 392, "role": "error_localization"}}
{"state": {"task": "Move the corn to the stove, then place it into the pot on the stove, then move it back to the counter top.", "plan": [{"step": 1, "subtask": "grasp corn"}, {"step": 2, "subtask": "move grasped corn to the stove"}, {"step": 3, "subtask": "grasp corn"}, {"step": 4, "subtask": "grasp corn"}, {"step": 5, "subtask": "move grasped corn to the pot on the stove"}, {"step": 6, "subtask": "release corn"}, {"step": 7, "subtask": "release corn"}, {"step": 8, "subtask": "move grasped corn to the counter top"}, {"step": 9, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d52daf007d42b664fe0709bd.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0362/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 362, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_23_train", "episode_id": "2", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 391, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the orange cloth from the right side of the table, then place it on the green towel, then move it back to the orange cloth", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the orange cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "move grasped silver pot to the green towel"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to the orange cloth"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3928ce5b24db942b387b9579.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0363/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 363, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_28_train", "episode_id": "13", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 422, "role": "error_localization"}}
{"state": {"task": "Move the red object to the left burner, then to the right burner, then back to the left burner.", "plan": [{"step": 1, "subtask": "grasp red object"}, {"step": 2, "subtask": "move grasped red object to left burner"}, {"step": 3, "subtask": "release red object"}, {"step": 4, "subtask": "grasp red object"}, {"step": 5, "subtask": "release red object"}, {"step": 6, "subtask": "grasp red object"}, {"step": 7, "subtask": "release red object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9d8611db6f5ba9488cf9e6b4.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0368/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_04_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 368, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_04_train", "episode_id": "17", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "put the blue fork on the orange towel, then put the red pepper in the pot, then put the pot in the sink", "plan": [{"step": 1, "subtask": "grasp blue fork"}, {"step": 2, "subtask": "move grasped blue fork to top of orange towel"}, {"step": 3, "subtask": "release blue fork"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to the pot"}, {"step": 6, "subtask": "release red pepper"}, {"step": 7, "subtask": "grasp red pepper"}, {"step": 8, "subtask": "move grasped object to the sink"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4d361addd4233dde18166a4f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0370/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_107_train/5", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 370, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_107_train", "episode_id": "5", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 395, "role": "error_localization"}}
{"state": {"task": "Move the purple towel beside the blue cup, then under the cup, then on top of it", "plan": [{"step": 1, "subtask": "grasp purple towel"}, {"step": 2, "subtask": "move grasped purple towel beside the blue cup"}, {"step": 3, "subtask": "release purple towel"}, {"step": 4, "subtask": "grasp purple towel"}, {"step": 5, "subtask": "move grasped purple towel on top of the blue cup"}, {"step": 6, "subtask": "release purple towel"}, {"step": 7, "subtask": "grasp purple towel"}, {"step": 8, "subtask": "move grasped purple towel beside the blue cup"}, {"step": 9, "subtask": "release purple towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/64864b2765e3815eec8d1db6.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0372/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 372, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train", "episode_id": "8", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 395, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the bottom left side of the stove, then the black pan to the center burner, and finally the silver pot to the top left side.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom left side of the stove"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped black pan to the center burner"}, {"step": 6, "subtask": "release black pan"}, {"step": 7, "subtask": "grasp black pan"}, {"step": 8, "subtask": "move grasped silver pot to the top left side of the stove"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bc4e76630abe1a643f2a2624.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0377/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_80_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 377, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_80_train", "episode_id": "13", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 414, "role": "error_localization"}}
{"state": {"task": "grab the yellow brush and put it beneath the silver pot on the left burner, then the silver pot on the right burner, then return the yellow brush to its original place", "plan": [{"step": 1, "subtask": "grasp yellow brush"}, {"step": 2, "subtask": "move grasped yellow brush beneath the silver pot on the right burner"}, {"step": 3, "subtask": "release yellow brush"}, {"step": 4, "subtask": "grasp yellow brush"}, {"step": 5, "subtask": "move grasped yellow brush to its original place"}, {"step": 6, "subtask": "release yellow brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9d923360a78c057154f29dc5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0379/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_44_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 379, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_44_train", "episode_id": "17", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "Remove the green object from the yellow cloth between the burners, then place it on the left burner, then move it to the right burner, then return it to the yellow cloth.", "plan": [{"step": 1, "subtask": "grasp green object"}, {"step": 2, "subtask": "move grasped green object out of yellow cloth"}, {"step": 3, "subtask": "release green object"}, {"step": 4, "subtask": "grasp green object"}, {"step": 5, "subtask": "move grasped green object to left burner"}, {"step": 6, "subtask": "release green object"}, {"step": 7, "subtask": "grasp green object"}, {"step": 8, "subtask": "move grasped green object to right burner"}, {"step": 9, "subtask": "release green object"}, {"step": 10, "subtask": "grasp green object"}, {"step": 11, "subtask": "move grasped green object to yellow cloth"}, {"step": 12, "subtask": "move it back to the original position"}, {"step": 13, "subtask": "release green object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/6ae788279934be2098cafe90.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0380/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 380, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_12_train", "episode_id": "17", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 482, "role": "error_localization"}}
{"state": {"task": "move the blue towel to the bottom left of the stove, then the red towel from the drawer next to it, then the green towel from the box", "plan": [{"step": 1, "subtask": "grasp blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the top right of the stove"}, {"step": 3, "subtask": "release blue towel"}, {"step": 4, "subtask": "grasp drawer handle"}, {"step": 5, "subtask": "move grasped drawer handle to drawer's closed position"}, {"step": 6, "subtask": "release drawer handle"}, {"step": 7, "subtask": "grasp red towel"}, {"step": 8, "subtask": "move grasped red towel out"}, {"step": 9, "subtask": "move grasped red towel to bottom left of the stove"}, {"step": 10, "subtask": "release red towel"}, {"step": 11, "subtask": "grasp green towel"}, {"step": 12, "subtask": "move grasped green towel out"}, {"step": 13, "subtask": "move grasped green towel to top right of the stove"}, {"step": 14, "subtask": "release green towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/983f41bd20a6a51e416e8cd0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0381/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 381, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_13_train", "episode_id": "17", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 505, "role": "error_localization"}}
{"state": {"task": "pick up the banana and put it between the burners, then remove it and place it on the countertop, then pick it up again and return it between the burners", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana between the burners"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped banana to countertop"}, {"step": 6, "subtask": "release banana"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "release banana"}, {"step": 9, "subtask": "move grasped banana between the burners"}, {"step": 10, "subtask": "release banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/125b5342fba4fb88cb682373.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0385/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 385, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_20_train", "episode_id": "8", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 414, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tap handle"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp bowl"}, {"step": 10, "subtask": "move grasped bowl to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp right burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner knob"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0388/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 388, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 609, "role": "error_localization"}}
{"state": {"task": "remove the knife from the top left from the table", "plan": [{"step": 1, "subtask": "pick up knife"}, {"step": 2, "subtask": "move grasped knife to the center of the table"}, {"step": 3, "subtask": "release knife"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c57f88c05cf053b13347a58a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0389/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train/21", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 389, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_07_train", "episode_id": "21", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 279, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up bowl"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up corn"}, {"step": 5, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd8f638fc68e4fd658fe55cb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0392/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/19", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 392, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "19", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 299, "role": "error_localization"}}
{"state": {"task": "move the red spoon to the front left corner of the stove, then to the front right corner, then to the back left corner, then to the back right corner", "plan": [{"step": 1, "subtask": "grasp red spoon"}, {"step": 2, "subtask": "move grasped red spoon to the front left corner of the stove"}, {"step": 3, "subtask": "release red spoon"}, {"step": 4, "subtask": "grasp red spoon"}, {"step": 5, "subtask": "move grasped red spoon to the front left corner of the stove"}, {"step": 6, "subtask": "release red spoon"}, {"step": 7, "subtask": "grasp red spoon"}, {"step": 8, "subtask": "move grasped red spoon to the back left corner of the stove"}, {"step": 9, "subtask": "release red spoon"}, {"step": 10, "subtask": "grasp red spoon"}, {"step": 11, "subtask": "move grasped red spoon to the back right corner of the stove"}, {"step": 12, "subtask": "release red spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ed39746680a92147fbfe4673.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0395/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_89_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 395, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_89_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 478, "role": "error_localization"}}
{"state": {"task": "Put the blue spoon to the left of the hot dog, then to its right, then above it, then below it.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the left of the hot dog"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "release blue spoon"}, {"step": 6, "subtask": "move grasped blue spoon to the right of the hot dog"}, {"step": 7, "subtask": "release blue spoon"}, {"step": 8, "subtask": "grasp blue spoon"}, {"step": 9, "subtask": "move grasped blue spoon above the hot dog"}, {"step": 10, "subtask": "release blue spoon"}, {"step": 11, "subtask": "grasp blue spoon"}, {"step": 12, "subtask": "move grasped blue spoon below the hot dog"}, {"step": 13, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c41861b6f740f9b3288edc2b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0396/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_100_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 396, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_100_train", "episode_id": "8", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 475, "role": "error_localization"}}
{"state": {"task": "Put the pot on top of the napkin, then place the pot on the table, then put it back on the napkin.", "plan": [{"step": 1, "subtask": "grasp pot"}, {"step": 2, "subtask": "move grasped pot to top of the napkin"}, {"step": 3, "subtask": "release pot"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "release pot"}, {"step": 6, "subtask": "release pot"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to top of the napkin"}, {"step": 9, "subtask": "move grasped pot to table"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/e5f6bd672ddccef0dda8196d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0397/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_00_train/16", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 397, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_00_train", "episode_id": "16", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 393, "role": "error_localization"}}
{"state": {"task": "cook the red pepper in the pot with the oven for 5 minutes, put the pot on the stove, close the oven, then remove the red pepper from the pot, put it on the table, put the pot in the sink", "plan": [{"step": 1, "subtask": "grasp oven door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to pot"}, {"step": 6, "subtask": "release red pepper"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to oven"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp oven door handle"}, {"step": 11, "subtask": "move grasped door handle to close position"}, {"step": 12, "subtask": "release door handle"}, {"step": 13, "subtask": "wait 5 minutes"}, {"step": 14, "subtask": "grasp oven door handle"}, {"step": 15, "subtask": "move grasped door handle to open position"}, {"step": 16, "subtask": "release door handle"}, {"step": 17, "subtask": "grasp yellow plate"}, {"step": 18, "subtask": "move grasped yellow plate to stove"}, {"step": 19, "subtask": "release yellow plate"}, {"step": 20, "subtask": "grasp oven door handle"}, {"step": 21, "subtask": "move grasped door handle to close position"}, {"step": 22, "subtask": "release door handle"}, {"step": 23, "subtask": "grasp red pepper"}, {"step": 24, "subtask": "move grasped red pepper to table"}, {"step": 25, "subtask": "release red pepper"}, {"step": 26, "subtask": "grasp pot"}, {"step": 27, "subtask": "move grasped pot to sink"}, {"step": 28, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b1c8d4924ab95f8f7484ab18.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0398/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_23_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 398, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_23_train", "episode_id": "2", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 752, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the bottom left of the burner, then to the top right, then back to the bottom left.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the top right of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the top right of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the top left of the burner"}, {"step": 9, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0bba7c89149c614833390126.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0399/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_14_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 399, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_14_train", "episode_id": "14", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 407, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from the blue cloth, place it on the wooden table, then return it to the cloth, finally move it to the green mat.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out of blue cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "move grasped silver pot to wooden table"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "grasp silver pot"}, {"step": 7, "subtask": "move grasped silver pot to blue cloth"}, {"step": 8, "subtask": "release silver pot"}, {"step": 9, "subtask": "grasp silver pot"}, {"step": 10, "subtask": "move grasped silver pot to green mat"}, {"step": 11, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8ce9b9ac4ef5ac0b94b36f71.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0401/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 401, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_21_train", "episode_id": "4", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 439, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom left side of the burner, then to the top right side, then back to the bottom left side, then to the top left side", "plan": [{"step": 1, "subtask": "move grasped silver pot to the bottom left side of the burner"}, {"step": 2, "subtask": "release silver pot"}, {"step": 3, "subtask": "move grasped silver pot to the top right side of the burner"}, {"step": 4, "subtask": "release silver pot"}, {"step": 5, "subtask": "grasp silver pot"}, {"step": 6, "subtask": "move grasped silver pot to the bottom left side of the burner"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to the top left side of the burner"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0ddc5fe40decc5c40c65f500.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0402/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_12_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 402, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_12_train", "episode_id": "7", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "open the drawer, take out each item one by one, place them on the table, then return them one by one into the drawer", "plan": [{"step": 1, "subtask": "grasp drawer handle"}, {"step": 2, "subtask": "move grasped drawer handle to drawer's open position"}, {"step": 3, "subtask": "release drawer handle"}, {"step": 4, "subtask": "grasp object"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to table"}, {"step": 7, "subtask": "release object"}, {"step": 8, "subtask": "move grasped object to drawer"}, {"step": 9, "subtask": "move grasped object backward"}, {"step": 10, "subtask": "release object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a029113e8089645841bb6d8b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0403/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train/26", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 403, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_20_train", "episode_id": "26", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 412, "role": "error_localization"}}
{"state": {"task": "remove the cucumber from the cup, then place it on the table, pick it up again, and put it back into the cup", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "release cucumber"}, {"step": 3, "subtask": "move grasped cucumber out of cup"}, {"step": 4, "subtask": "release cucumber"}, {"step": 5, "subtask": "grasp cucumber"}, {"step": 6, "subtask": "move grasped cucumber to cup"}, {"step": 7, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ca455acb6df19780a4ec69db.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0404/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 404, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "move the blue brush to upper side of the stove next to yellow cloth, then move it to lower side, then back to upper side", "plan": [{"step": 1, "subtask": "grasp blue brush"}, {"step": 2, "subtask": "move grasped blue brush to the lower side of the stove"}, {"step": 3, "subtask": "release blue brush"}, {"step": 4, "subtask": "grasp blue brush"}, {"step": 5, "subtask": "move grasped blue brush to the upper side of the stove"}, {"step": 6, "subtask": "release blue brush"}, {"step": 7, "subtask": "grasp blue brush"}, {"step": 8, "subtask": "move grasped blue brush to the lower side of the stove"}, {"step": 9, "subtask": "release blue brush"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0fbf5aaeb262929ab97488be.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0405/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_50_train/16", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 405, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_50_train", "episode_id": "16", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 410, "role": "error_localization"}}
{"state": {"task": "Move the silver pot on top of the green cloth, then place it on the blue mat, then return it to the cloth.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to blue mat"}, {"step": 6, "subtask": "bring it back to its original position"}, {"step": 7, "subtask": "release silver pot"}, {"step": 8, "subtask": "grasp silver pot"}, {"step": 9, "subtask": "move grasped silver pot to top of the green cloth"}, {"step": 10, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dfd421c8d0c3cbba5e2cc99.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0407/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_85_train/3", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 407, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_85_train", "episode_id": "3", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 422, "role": "error_localization"}}
{"state": {"task": "Move the red pepper on top of the orange cloth, then on the blue plate, then back to the orange cloth.", "plan": [{"step": 1, "subtask": "grasp red pepper"}, {"step": 2, "subtask": "move grasped red pepper to top of the orange cloth"}, {"step": 3, "subtask": "release red pepper"}, {"step": 4, "subtask": "grasp red pepper"}, {"step": 5, "subtask": "move grasped red pepper to blue plate"}, {"step": 6, "subtask": "release red pepper"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7461e053a5b6e69086656776.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0410/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 410, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_22_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 346, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the bottom right corner of the burner, then to the top left, then to the center", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom right corner of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the center of the burner"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c90750c2ccb5c44fcd5f9de5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0413/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_38_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 413, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_38_train", "episode_id": "10", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "Move the spoon to the left, then to the right, then to the top, then to the bottom.", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to the left"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon to the right"}, {"step": 6, "subtask": "release spoon"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon to the top"}, {"step": 9, "subtask": "release spoon"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon to the bottom"}, {"step": 12, "subtask": "move it back to the original position"}, {"step": 13, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/58dc6b4bac1e4a0fa4b8d4fb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0414/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_36_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 414, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_36_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "Move one of the cans to the upper left corner of the left burner, then to the upper right corner, then to the lower left corner, and finally to the lower right corner.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "move grasped can to the upper left corner of the left burner"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the lower left corner of the left burner"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the lower right corner of the left burner"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "move grasped can to the upper right corner of the left burner"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c850847225497b2c7a0b7727.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0415/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train/2", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 415, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_00_train", "episode_id": "2", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 473, "role": "error_localization"}}
{"state": {"task": "remove the green vegetable from the right side of the burner, then place it on the left side, then return it to the right side, then remove it again.", "plan": [{"step": 1, "subtask": "grasp green vegetable"}, {"step": 2, "subtask": "move grasped green vegetable out"}, {"step": 3, "subtask": "move grasped green vegetable to left side of the burner"}, {"step": 4, "subtask": "release green vegetable"}, {"step": 5, "subtask": "grasp green vegetable"}, {"step": 6, "subtask": "move grasped green vegetable out"}, {"step": 7, "subtask": "move grasped green vegetable to left side of the burner"}, {"step": 8, "subtask": "release green vegetable"}, {"step": 9, "subtask": "grasp green vegetable"}, {"step": 10, "subtask": "move grasped green vegetable out"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/ec5cb938ff4a1a745a46f539.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0418/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 418, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_combo_toykitchen6_03_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 429, "role": "error_localization"}}
{"state": {"task": "remove cup from plate", "plan": [{"step": 1, "subtask": "pick up cup"}, {"step": 2, "subtask": "move grasped cup away from plate"}, {"step": 3, "subtask": "place cup back on plate"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/a7a4e8599971151e571b8b49.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0420/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train/23", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 420, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_cup_on_plate_train", "episode_id": "23", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 273, "role": "error_localization"}}
{"state": {"task": "move the red fork to the left side of the stove, then to the right side, then back to the left, then place it on top of the stove", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "grasp red fork"}, {"step": 5, "subtask": "release red fork"}, {"step": 6, "subtask": "move grasped red fork to the right side of the stove"}, {"step": 7, "subtask": "release red fork"}, {"step": 8, "subtask": "grasp red fork"}, {"step": 9, "subtask": "move grasped red fork to the left side of the stove"}, {"step": 10, "subtask": "release red fork"}, {"step": 11, "subtask": "grasp red fork"}, {"step": 12, "subtask": "move grasped red fork to top of the stove"}, {"step": 13, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b9aec55453901a208bc241d7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0421/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_07_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 421, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_07_train", "episode_id": "11", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 487, "role": "error_localization"}}
{"state": {"task": "put red pot on the right burner, add the tomato and cheese into it, turn on the corresponding burner", "plan": [{"step": 1, "subtask": "grasp red tomato"}, {"step": 2, "subtask": "move grasped red tomato to right burner"}, {"step": 3, "subtask": "release red tomato"}, {"step": 4, "subtask": "grasp tomato"}, {"step": 5, "subtask": "move grasped tomato into red pot"}, {"step": 6, "subtask": "release tomato"}, {"step": 7, "subtask": "grasp cheese"}, {"step": 8, "subtask": "move grasped cheese into red pot"}, {"step": 9, "subtask": "release cheese"}, {"step": 10, "subtask": "grasp right burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0422/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 422, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the right bottom of the burner, then to the left top, then to the center, then back to the right bottom", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the right bottom of the burner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the right bottom of the burner"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the center of the burner"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the left top of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5c33dae017f4a15ba5e162f9.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0424/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_56_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 424, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_56_train", "episode_id": "14", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 469, "role": "error_localization"}}
{"state": {"task": "put the cans in the cupboard, put the bread on the table, and put the red towel, and the orange bowl in the sink", "plan": [{"step": 1, "subtask": "grasp cupboard door handle"}, {"step": 2, "subtask": "move grasped door handle to open position"}, {"step": 3, "subtask": "release door handle"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to cupboard"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp another can"}, {"step": 8, "subtask": "move grasped can to cupboard"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp croissant"}, {"step": 11, "subtask": "move grasped croissant to table"}, {"step": 12, "subtask": "release croissant"}, {"step": 13, "subtask": "grasp red towel"}, {"step": 14, "subtask": "move grasped red towel to sink"}, {"step": 15, "subtask": "release red towel"}, {"step": 16, "subtask": "grasp orange bowl"}, {"step": 17, "subtask": "move grasped orange bowl to sink"}, {"step": 18, "subtask": "release orange bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3a3e49830f9cc5aa16c21e62.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0425/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 425, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_01_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 551, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the vegetable can", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp brocoli"}, {"step": 11, "subtask": "move grasped brocoli to upper right corner"}, {"step": 12, "subtask": "release brocoli"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0426/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 426, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 459, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the bottom right corner to the top right corner, then to the bottom left corner, then to the top left corner, and finally put to the upper right corner the vegetable", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to top right corner"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to bottom left corner"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to top left corner"}, {"step": 9, "subtask": "release cloth"}, {"step": 10, "subtask": "grasp tomato"}, {"step": 11, "subtask": "move grasped tomato to upper right corner"}, {"step": 12, "subtask": "release tomato"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/c49ff2ae4db094aeff42269a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0430/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train/14", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 430, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_27_train", "episode_id": "14", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 455, "role": "error_localization"}}
{"state": {"task": "Move the blue object to the top right spot of the right burner, then to the center, then to the bottom left spot, then back to the top right spot.", "plan": [{"step": 1, "subtask": "grasp blue object"}, {"step": 2, "subtask": "move grasped blue object to the top right spot of the right burner"}, {"step": 3, "subtask": "release blue object"}, {"step": 4, "subtask": "grasp blue object"}, {"step": 5, "subtask": "move grasped blue object to the center"}, {"step": 6, "subtask": "release blue object"}, {"step": 7, "subtask": "grasp blue object"}, {"step": 8, "subtask": "grasp blue object"}, {"step": 9, "subtask": "release blue object"}, {"step": 10, "subtask": "move grasped blue object to the bottom left spot"}, {"step": 11, "subtask": "move grasped blue object to the top right spot of the right burner"}, {"step": 12, "subtask": "release blue object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/4abc430ddbc71789b2a7473a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0431/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_46_train/9", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 431, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_46_train", "episode_id": "9", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 472, "role": "error_localization"}}
{"state": {"task": "move the blue cup to the bottom left of the right burner, then to the top left, then to the bottom right, then back to the center of the right burner", "plan": [{"step": 1, "subtask": "grasp blue cup"}, {"step": 2, "subtask": "move grasped blue cup to the top left of the right burner"}, {"step": 3, "subtask": "release blue cup"}, {"step": 4, "subtask": "grasp blue cup"}, {"step": 5, "subtask": "move grasped blue cup to the top left of the right burner"}, {"step": 6, "subtask": "release blue cup"}, {"step": 7, "subtask": "grasp blue cup"}, {"step": 8, "subtask": "move grasped blue cup to the bottom left of the right burner"}, {"step": 9, "subtask": "release blue cup"}, {"step": 10, "subtask": "grasp blue cup"}, {"step": 11, "subtask": "move grasped blue cup to the top left of the right burner"}, {"step": 12, "subtask": "release blue cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d43ef9cab7d5a642a7e06349.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0432/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_56_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 432, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_56_train", "episode_id": "10", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 479, "role": "error_localization"}}
{"state": {"task": "Move the cloth to bottom from top, then to the left, then to the right, then back to the bottom.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to left"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth to right"}, {"step": 6, "subtask": "release cloth"}, {"step": 7, "subtask": "grasp cloth"}, {"step": 8, "subtask": "move grasped cloth to bottom"}, {"step": 9, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1af6cec5704137455abc579e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0434/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 434, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_05_train", "episode_id": "13", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 382, "role": "error_localization"}}
{"state": {"task": "Move the can to the lower right of burner, then to the upper left, then to the center, then back to the lower right.", "plan": [{"step": 1, "subtask": "move grasped can to the lower right of burner"}, {"step": 2, "subtask": "move grasped can to the lower right of burner"}, {"step": 3, "subtask": "release can"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped can to the upper left"}, {"step": 6, "subtask": "release can"}, {"step": 7, "subtask": "grasp can"}, {"step": 8, "subtask": "move grasped can to the center"}, {"step": 9, "subtask": "release can"}, {"step": 10, "subtask": "grasp can"}, {"step": 11, "subtask": "grasp can"}, {"step": 12, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/295587a23223a9f08e2e105b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0435/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 435, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_25_train", "episode_id": "1", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 447, "role": "error_localization"}}
{"state": {"task": "put the cup to the counter from the sink, then to the stove, then back to the sink, then to the counter again", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to counter"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp cup"}, {"step": 5, "subtask": "move grasped cup to stove"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to sink"}, {"step": 9, "subtask": "release cup"}, {"step": 10, "subtask": "grasp cup"}, {"step": 11, "subtask": "move grasped cup to counter"}, {"step": 12, "subtask": "put it back where it was"}, {"step": 13, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/23e98788988f09c9c5829a59.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0436/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train/29", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 436, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_counter_to_sink_train", "episode_id": "29", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "move the cloth from the left burner to the right burner", "plan": [{"step": 1, "subtask": "pick up cloth"}, {"step": 2, "subtask": "move grasped cloth to the left burner"}, {"step": 3, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0475b18118c518a6e66c64b7.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0440/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 440, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_10_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 278, "role": "error_localization"}}
{"state": {"task": "remove spoon from bowl, then place spoon on countertop, then put spoon back in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon out of bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "release spoon"}, {"step": 6, "subtask": "move grasped spoon to countertop"}, {"step": 7, "subtask": "release spoon"}, {"step": 8, "subtask": "grasp spoon"}, {"step": 9, "subtask": "move grasped spoon to bowl"}, {"step": 10, "subtask": "release spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f9951d0fcde3b0635648233f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0441/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/28", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 441, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "28", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 394, "role": "error_localization"}}
{"state": {"task": "Move the silver pot to the left side of the burner, then to the front, then to the right side, and finally back to the left.", "plan": [{"step": 1, "subtask": "release silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left side of the burner"}, {"step": 3, "subtask": "grasp silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the front"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "grasp silver pot"}, {"step": 8, "subtask": "move grasped silver pot to the right side of the burner"}, {"step": 9, "subtask": "release silver pot"}, {"step": 10, "subtask": "grasp silver pot"}, {"step": 11, "subtask": "move grasped silver pot to the left side of the burner"}, {"step": 12, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/8f970e9600791f7a83997e2b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0442/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_57_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 442, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_57_train", "episode_id": "13", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 466, "role": "error_localization"}}
{"state": {"task": "heat the orange in the microwave then put it in the pot, and put the silver lid on the pot, then put the pot on the bottom stove and turn it on", "plan": [{"step": 1, "subtask": "grasp orange"}, {"step": 2, "subtask": "move grasped orange to microwave"}, {"step": 3, "subtask": "release orange"}, {"step": 4, "subtask": "grasp orange"}, {"step": 5, "subtask": "move grasped orange out of microwave"}, {"step": 6, "subtask": "move grasped orange to the pot"}, {"step": 7, "subtask": "release orange"}, {"step": 8, "subtask": "grasp silver lid"}, {"step": 9, "subtask": "move silver lid to top of pot"}, {"step": 10, "subtask": "release silver lid"}, {"step": 11, "subtask": "grasp pot"}, {"step": 12, "subtask": "move grasped silver pot to the upper stove"}, {"step": 13, "subtask": "release pot"}, {"step": 14, "subtask": "grasp upper stove knob"}, {"step": 15, "subtask": "turn it on"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/113282c5a87d7796e1031a58.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0449/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train/6", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 449, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_26_train", "episode_id": "6", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 518, "role": "error_localization"}}
{"state": {"task": "move the yellow towel on the upper left corner of the stove, then fold it and place it on the upper right corner, then unfold it and return it to the upper left corner", "plan": [{"step": 1, "subtask": "grasp yellow towel"}, {"step": 2, "subtask": "move grasped yellow towel to the upper right corner of the stove"}, {"step": 3, "subtask": "release yellow towel"}, {"step": 4, "subtask": "grasp yellow towel"}, {"step": 5, "subtask": "release yellow towel"}, {"step": 6, "subtask": "move grasped yellow towel to the upper left corner of the stove"}, {"step": 7, "subtask": "rotate grasped yellow towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52949f18fe35fc8c79927aad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0452/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_94_train/12", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 452, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_94_train", "episode_id": "12", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 383, "role": "error_localization"}}
{"state": {"task": "Put the yellow doll at the center of the table, then move it to the front left corner, then to the back right corner, finally return it to the center.", "plan": [{"step": 1, "subtask": "grasp yellow doll"}, {"step": 2, "subtask": "move grasped yellow doll to the center of the table"}, {"step": 3, "subtask": "release yellow doll"}, {"step": 4, "subtask": "grasp yellow doll"}, {"step": 5, "subtask": "move grasped yellow doll to the front left corner"}, {"step": 6, "subtask": "release yellow doll"}, {"step": 7, "subtask": "grasp yellow doll"}, {"step": 8, "subtask": "move grasped yellow doll to the front left corner"}, {"step": 9, "subtask": "release yellow doll"}, {"step": 10, "subtask": "grasp yellow doll"}, {"step": 11, "subtask": "move grasped yellow doll to the center of the table"}, {"step": 12, "subtask": "release yellow doll"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/223ec694abb51b4ba5e33564.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0454/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train/24", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 454, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_16_train", "episode_id": "24", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 468, "role": "error_localization"}}
{"state": {"task": "Move the purple cloth to the bottom right corner of the stove, then fold it, then unfold it, then move it to the bottom left corner of the stove.", "plan": [{"step": 1, "subtask": "grasp purple cloth"}, {"step": 2, "subtask": "move grasped purple cloth to the bottom right corner of the stove"}, {"step": 3, "subtask": "release purple cloth"}, {"step": 4, "subtask": "grasp purple cloth"}, {"step": 5, "subtask": "move grasped purple cloth to folded position"}, {"step": 6, "subtask": "bring it back to its original position"}, {"step": 7, "subtask": "release purple cloth"}, {"step": 8, "subtask": "grasp purple cloth"}, {"step": 9, "subtask": "move grasped purple cloth to unfolded position"}, {"step": 10, "subtask": "release purple cloth"}, {"step": 11, "subtask": "grasp purple cloth"}, {"step": 12, "subtask": "move grasped purple cloth to the bottom left corner of the stove"}, {"step": 13, "subtask": "release purple cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/188043555a15ef77efd456e0.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0455/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_106_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 455, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_106_train", "episode_id": "1", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 487, "role": "error_localization"}}
{"state": {"task": "Move the yellow banana between the silver pot and the blue cloth, then place it on the silver pot, then return it to the original position.", "plan": [{"step": 1, "subtask": "grasp yellow banana"}, {"step": 2, "subtask": "move grasped yellow banana between the silver pot and the blue cloth"}, {"step": 3, "subtask": "release yellow banana"}, {"step": 4, "subtask": "grasp yellow banana"}, {"step": 5, "subtask": "move grasped yellow banana to the silver pot"}, {"step": 6, "subtask": "release yellow banana"}, {"step": 7, "subtask": "grasp yellow banana"}, {"step": 8, "subtask": "move grasped yellow banana to original position"}, {"step": 9, "subtask": "release yellow banana"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/099b6feec710b502144f656f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0458/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 458, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_05_train", "episode_id": "13", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 405, "role": "error_localization"}}
{"state": {"task": "Remove the silver pot from top of the green cloth, place it on the table, then return it to the cloth, finally lift the cloth to reveal what's underneath.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot out"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "grasp green cloth"}, {"step": 6, "subtask": "release silver pot"}, {"step": 7, "subtask": "move grasped silver pot to green cloth"}, {"step": 8, "subtask": "move grasped green cloth up"}, {"step": 9, "subtask": "release green cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/63fcd9930f4272aa1004534d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0459/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_105_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 459, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_105_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 400, "role": "error_localization"}}
{"state": {"task": "Move the cloth from the edge of the table to the top, then fold it neatly, then place it on the chair.", "plan": [{"step": 1, "subtask": "move grasped cloth to the top of the table"}, {"step": 2, "subtask": "grasp cloth"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "move grasped cloth forward"}, {"step": 6, "subtask": "move grasped cloth to the left"}, {"step": 7, "subtask": "move grasped cloth to the right"}, {"step": 8, "subtask": "release cloth"}, {"step": 9, "subtask": "grasp cloth"}, {"step": 10, "subtask": "move grasped cloth to the chair"}, {"step": 11, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0043486c552bff5e13c89b03.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0460/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 460, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_pnp_15_train", "episode_id": "4", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 427, "role": "error_localization"}}
{"state": {"task": "Take the red fork and place it on the green cloth, then remove it and place it on the blue plate, then return it to the green cloth.", "plan": [{"step": 1, "subtask": "grasp red fork"}, {"step": 2, "subtask": "move grasped red fork to green cloth"}, {"step": 3, "subtask": "release red fork"}, {"step": 4, "subtask": "release red fork"}, {"step": 5, "subtask": "grasp red fork"}, {"step": 6, "subtask": "move grasped red fork to green cloth"}, {"step": 7, "subtask": "release red fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/91cde32bc18d285d1f748739.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0461/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 461, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_37_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 365, "role": "error_localization"}}
{"state": {"task": "Move the blue spoon to the right burner, then to the left burner, then back to the right burner.", "plan": [{"step": 1, "subtask": "grasp blue spoon"}, {"step": 2, "subtask": "move grasped blue spoon to the left burner"}, {"step": 3, "subtask": "release blue spoon"}, {"step": 4, "subtask": "grasp blue spoon"}, {"step": 5, "subtask": "move grasped blue spoon to the right burner"}, {"step": 6, "subtask": "release blue spoon"}, {"step": 7, "subtask": "grasp blue spoon"}, {"step": 8, "subtask": "move grasped blue spoon to the right burner"}, {"step": 9, "subtask": "release blue spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b6d56a60cccc6ceb77269f32.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0464/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train/6", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 464, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_08_train", "episode_id": "6", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 395, "role": "error_localization"}}
{"state": {"task": "move blue doll from top of the drawer and place it on the table", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up blue doll"}, {"step": 3, "subtask": "move grasped blue doll to drawer"}, {"step": 4, "subtask": "release blue doll"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1ad6671a2052a36849b18cad.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0466/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 466, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "7", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 296, "role": "error_localization"}}
{"state": {"task": "move the blue towel to the upper left of burner, then to the upper right of burner, then to the lower left of burner, then to the lower right of burner", "plan": [{"step": 1, "subtask": "grasp blue towel"}, {"step": 2, "subtask": "move grasped blue towel to the upper left of burner"}, {"step": 3, "subtask": "release blue towel"}, {"step": 4, "subtask": "grasp blue towel"}, {"step": 5, "subtask": "move grasped blue towel to the lower left of burner"}, {"step": 6, "subtask": "release blue towel"}, {"step": 7, "subtask": "grasp blue towel"}, {"step": 8, "subtask": "move grasped blue towel to the lower left of burner"}, {"step": 9, "subtask": "release blue towel"}, {"step": 10, "subtask": "grasp blue towel"}, {"step": 11, "subtask": "move grasped blue towel to the lower right of burner"}, {"step": 12, "subtask": "release blue towel"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/5a1078e2668ddfbbb23fbbbc.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0467/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train/11", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 467, "taskvar": "numpy_256_bridge_data_v2_datacol2_toykitchen6_many_skills_37_train", "episode_id": "11", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 471, "role": "error_localization"}}
{"state": {"task": "put red pot on the right burner, add the tomato and cheese into it, turn on the corresponding burner", "plan": [{"step": 1, "subtask": "grasp red pot"}, {"step": 2, "subtask": "move grasped red pot to right burner"}, {"step": 3, "subtask": "release red pot"}, {"step": 4, "subtask": "grasp tomato"}, {"step": 5, "subtask": "move grasped tomato into red pot"}, {"step": 6, "subtask": "release tomato"}, {"step": 7, "subtask": "grasp yellow peach"}, {"step": 8, "subtask": "move grasped yellow peach into red pot"}, {"step": 9, "subtask": "release yellow peach"}, {"step": 10, "subtask": "grasp right burner knob"}, {"step": 11, "subtask": "rotate knob to ON position"}, {"step": 12, "subtask": "release knob"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/08feed7f24e4d18e0149b194.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0468/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train/21", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 468, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_beet_in_pot_sink_train", "episode_id": "21", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 440, "role": "error_localization"}}
{"state": {"task": "place the blue fork below the right burner, then to the left burner, then back below the right burner", "plan": [{"step": 1, "subtask": "grasp blue fork"}, {"step": 2, "subtask": "move grasped blue fork below the right burner"}, {"step": 3, "subtask": "put it back where it was"}, {"step": 4, "subtask": "release blue fork"}, {"step": 5, "subtask": "grasp blue fork"}, {"step": 6, "subtask": "move grasped blue fork to the left burner"}, {"step": 7, "subtask": "release blue fork"}, {"step": 8, "subtask": "grasp blue fork"}, {"step": 9, "subtask": "move grasped blue fork below the right burner"}, {"step": 10, "subtask": "release blue fork"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/3bdf5023e0932785300f5e01.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0469/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_107_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 469, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_107_train", "episode_id": "4", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 414, "role": "error_localization"}}
{"state": {"task": "put the cucumber in the cup, then remove the cucumber, then put it back in, then tilt the cup to roll out the cucumber", "plan": [{"step": 1, "subtask": "grasp cucumber"}, {"step": 2, "subtask": "move grasped cucumber to cup"}, {"step": 3, "subtask": "release cucumber"}, {"step": 4, "subtask": "grasp cucumber"}, {"step": 5, "subtask": "move grasped cucumber out"}, {"step": 6, "subtask": "release cucumber"}, {"step": 7, "subtask": "grasp cucumber"}, {"step": 8, "subtask": "move grasped cucumber to table"}, {"step": 9, "subtask": "release cucumber"}, {"step": 10, "subtask": "grasp cucumber"}, {"step": 11, "subtask": "move grasped cucumber forward"}, {"step": 12, "subtask": "rotate grasped cucumber"}, {"step": 13, "subtask": "release cucumber"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/287cb4f6a2dca96741ac6d87.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0470/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train/22", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 470, "taskvar": "numpy_256_bridge_data_v1_berkeley_tabletop_light_wood_put_cucumber_in_cup_train", "episode_id": "22", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 451, "role": "error_localization"}}
{"state": {"task": "move the white and orange object on the right burner", "plan": [{"step": 1, "subtask": "pick up white object"}, {"step": 2, "subtask": "move grasped white object to right burner"}, {"step": 3, "subtask": "release white object"}, {"step": 4, "subtask": "move grasped orange object to right burner"}, {"step": 5, "subtask": "release orange object"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/d2041f4c15a4c9033cbdba7d.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0472/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train/12", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 472, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_toykitchen6_01_train", "episode_id": "12", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 314, "role": "error_localization"}}
{"state": {"task": "Place the capsicum near the ladle, then move the ladle to the center of the table, then place the capsicum next to the ladle again.", "plan": [{"step": 1, "subtask": "grasp capsicum"}, {"step": 2, "subtask": "move grasped capsicum near the ladle"}, {"step": 3, "subtask": "release capsicum"}, {"step": 4, "subtask": "grasp capsicum"}, {"step": 5, "subtask": "move grasped ladle to the center of the table"}, {"step": 6, "subtask": "release ladle"}, {"step": 7, "subtask": "grasp ladle"}, {"step": 8, "subtask": "move grasped capsicum next to the ladle"}, {"step": 9, "subtask": "release capsicum"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/96837ab6c4383dc63e004864.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0474/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_104_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 474, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_104_train", "episode_id": "10", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 409, "role": "error_localization"}}
{"state": {"task": "fill bowl with water, then put the bowl on the left burner, turn it on and heat the bowl for 5 minutes then turn it off and then put the bowl on the table", "plan": [{"step": 1, "subtask": "grasp bowl"}, {"step": 2, "subtask": "move grasped bowl under the faucet"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp oven handle"}, {"step": 5, "subtask": "rotate grasped handle to open position"}, {"step": 6, "subtask": "wait for the bowl to be filled"}, {"step": 7, "subtask": "rotate grasped handle to closed position"}, {"step": 8, "subtask": "release handle"}, {"step": 9, "subtask": "grasp bowl"}, {"step": 10, "subtask": "move grasped bowl to left burner"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp left burner knob"}, {"step": 13, "subtask": "rotate grasped knob to on position"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "wait 5 minutes"}, {"step": 16, "subtask": "grasp left burner knob"}, {"step": 17, "subtask": "rotate grasped knob to off position"}, {"step": 18, "subtask": "release knob"}, {"step": 19, "subtask": "grasp bowl"}, {"step": 20, "subtask": "move grasped bowl to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/95463bc3d4e47e6356af3c79.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0475/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/17", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 475, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "17", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 609, "role": "error_localization"}}
{"state": {"task": "take corn out of bowl", "plan": [{"step": 1, "subtask": "pick up corn"}, {"step": 2, "subtask": "move grasped bowl to sink"}, {"step": 3, "subtask": "release bowl"}, {"step": 4, "subtask": "pick up bowl"}, {"step": 5, "subtask": "move grasped corn to sink"}, {"step": 6, "subtask": "release corn"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/bd8f638fc68e4fd658fe55cb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0479/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train/19", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 479, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_take_corn_out_of_bowl_sink_train", "episode_id": "19", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 317, "role": "error_localization"}}
{"state": {"task": "move the silver pot to the left, then to the right, then to the center, then back to the left.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the left"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "release silver pot"}, {"step": 6, "subtask": "grasp silver pot"}, {"step": 7, "subtask": "move grasped silver pot to the center"}, {"step": 8, "subtask": "release silver pot"}, {"step": 9, "subtask": "grasp silver pot"}, {"step": 10, "subtask": "move grasped silver pot to the left"}, {"step": 11, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2eafcc3f152844874a2521a5.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0480/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train/4", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 480, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_pnp_sweep_04_train", "episode_id": "4", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 427, "role": "error_localization"}}
{"state": {"task": "move the silver pot at the top left corner of the burner, then to the top right corner, then to the bottom left corner, then to the bottom right corner.", "plan": [{"step": 1, "subtask": "grasp silver pot"}, {"step": 2, "subtask": "move grasped silver pot to the bottom left corner"}, {"step": 3, "subtask": "release silver pot"}, {"step": 4, "subtask": "grasp silver pot"}, {"step": 5, "subtask": "move grasped silver pot to the bottom right corner"}, {"step": 6, "subtask": "release silver pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/1dc6356b27c036dfe2829225.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0481/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_58_train/6", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 481, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_58_train", "episode_id": "6", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 357, "role": "error_localization"}}
{"state": {"task": "Remove the blue cloth from the top edge of the cloth, then place it on the bottom edge, then return it to the top edge, then set it aside.", "plan": [{"step": 1, "subtask": "move grasped blue cloth away from the top edge of the cloth"}, {"step": 2, "subtask": "release blue cloth"}, {"step": 3, "subtask": "move grasped blue cloth to the bottom edge of the cloth"}, {"step": 4, "subtask": "release blue cloth"}, {"step": 5, "subtask": "grasp blue cloth"}, {"step": 6, "subtask": "move grasped blue cloth to the top edge of the cloth"}, {"step": 7, "subtask": "release blue cloth"}, {"step": 8, "subtask": "grasp blue cloth"}, {"step": 9, "subtask": "move grasped blue cloth away from the top edge of the cloth"}, {"step": 10, "subtask": "release blue cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/86116ee6494d87d8d320afb2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0484/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train/8", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 484, "taskvar": "numpy_256_bridge_data_v2_datacol1_toykitchen6_fold_cloth_02_train", "episode_id": "8", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 441, "role": "error_localization"}}
{"state": {"task": "Move the mushroom to the lower right of the cooktop, then to the upper left, then back to the lower right, then to the center.", "plan": [{"step": 1, "subtask": "grasp mushroom"}, {"step": 2, "subtask": "move grasped mushroom to the lower right of the cooktop"}, {"step": 3, "subtask": "release mushroom"}, {"step": 4, "subtask": "move grasped mushroom to the upper left of the cooktop"}, {"step": 5, "subtask": "release mushroom"}, {"step": 6, "subtask": "grasp mushroom"}, {"step": 7, "subtask": "release mushroom"}, {"step": 8, "subtask": "grasp mushroom"}, {"step": 9, "subtask": "move grasped mushroom to the center of the cooktop"}, {"step": 10, "subtask": "release mushroom"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/2dac1307576affb964ab9ef8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0485/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_95_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 485, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_95_train", "episode_id": "1", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 420, "role": "error_localization"}}
{"state": {"task": "move the cheese to the right of the blue brush, then to the left of the green cup, then back to the right of the blue brush.", "plan": [{"step": 1, "subtask": "grasp cheese"}, {"step": 2, "subtask": "move grasped cheese to the left of the green cup"}, {"step": 3, "subtask": "release cheese"}, {"step": 4, "subtask": "grasp cheese"}, {"step": 5, "subtask": "move grasped cheese to the right of the blue brush"}, {"step": 6, "subtask": "release cheese"}, {"step": 7, "subtask": "grasp cheese"}, {"step": 8, "subtask": "move grasped cheese to the left of the green cup"}, {"step": 9, "subtask": "release cheese"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f8ecd81906d7736b2c835391.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0487/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_91_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 487, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_91_train", "episode_id": "13", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 403, "role": "error_localization"}}
{"state": {"task": "remove the yellow duck from the drawer", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "move grasped yellow duck away from drawer"}, {"step": 3, "subtask": "release yellow duck"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/cadf3573d63e2515c5e0acd8.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0488/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train/30", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 488, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_25_train", "episode_id": "30", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 274, "role": "error_localization"}}
{"state": {"task": "move cloth to top from bottom, then fold it, then place it to the side.", "plan": [{"step": 1, "subtask": "grasp cloth"}, {"step": 2, "subtask": "move grasped cloth to side"}, {"step": 3, "subtask": "release cloth"}, {"step": 4, "subtask": "grasp cloth"}, {"step": 5, "subtask": "rotate grasped cloth"}, {"step": 6, "subtask": "push down"}, {"step": 7, "subtask": "release cloth"}, {"step": 8, "subtask": "grasp cloth"}, {"step": 9, "subtask": "move grasped cloth to top"}, {"step": 10, "subtask": "release cloth"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/fbd5256059f152bb7404ce4b.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0490/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train/1", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 490, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_fold_cloth_17_train", "episode_id": "1", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 389, "role": "error_localization"}}
{"state": {"task": "Move the yellow spoon to the upper centre of the table, then to the lower left corner, then back to the upper centre, then to the upper right corner.", "plan": [{"step": 1, "subtask": "grasp yellow spoon"}, {"step": 2, "subtask": "move grasped yellow spoon to the upper centre of the table"}, {"step": 3, "subtask": "release yellow spoon"}, {"step": 4, "subtask": "grasp yellow spoon"}, {"step": 5, "subtask": "move grasped yellow spoon to the lower left corner of the table"}, {"step": 6, "subtask": "release yellow spoon"}, {"step": 7, "subtask": "grasp yellow spoon"}, {"step": 8, "subtask": "move grasped yellow spoon to the upper centre of the table"}, {"step": 9, "subtask": "release yellow spoon"}, {"step": 10, "subtask": "grasp yellow spoon"}, {"step": 11, "subtask": "release yellow spoon"}, {"step": 12, "subtask": "move grasped yellow spoon to the upper right corner of the table"}, {"step": 13, "subtask": "release yellow spoon"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/52960557c3d47f6f695b3e8e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0491/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train/20", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 491, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_many_skills_13_train", "episode_id": "20", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 491, "role": "error_localization"}}
{"state": {"task": "put spatula on plate sink, then move plate to drying rack, then move spatula to cutting board", "plan": [{"step": 1, "subtask": "grasp spatula"}, {"step": 2, "subtask": "move grasped spatula to plate"}, {"step": 3, "subtask": "release spatula"}, {"step": 4, "subtask": "grasp spatula"}, {"step": 5, "subtask": "move grasped spatula to cutting board"}, {"step": 6, "subtask": "release spatula"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/f10a3cb9e75f3eacd9db109f.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0494/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train/40", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 494, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spatula_on_plate_sink_train", "episode_id": "40", "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 340, "role": "error_localization"}}
{"state": {"task": "put cup from drying rack to sink, then take another cup from the drying rack to the shelf, and then move it to the sink", "plan": [{"step": 1, "subtask": "grasp cup"}, {"step": 2, "subtask": "move grasped cup to shelf"}, {"step": 3, "subtask": "release cup"}, {"step": 4, "subtask": "grasp another cup"}, {"step": 5, "subtask": "move grasped cup to sink"}, {"step": 6, "subtask": "release cup"}, {"step": 7, "subtask": "grasp cup"}, {"step": 8, "subtask": "move grasped cup to drying rack"}, {"step": 9, "subtask": "release cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/b669d05bfc254c40877b9b34.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0495/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train/9", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 495, "taskvar": "numpy_256_bridge_data_v1_upenn_toysink4_put_cup_from_drying_rack_to_sink_train", "episode_id": "9", "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 388, "role": "error_localization"}}
{"state": {"task": "put the pot on the blue cloth, then move it to the right side of the cloth, then to the left side, then back to the center", "plan": [{"step": 1, "subtask": "grasp pot"}, {"step": 2, "subtask": "move grasped pot to blue cloth"}, {"step": 3, "subtask": "release pot"}, {"step": 4, "subtask": "grasp pot"}, {"step": 5, "subtask": "release pot"}, {"step": 6, "subtask": "move grasped pot to right side of blue cloth"}, {"step": 7, "subtask": "grasp pot"}, {"step": 8, "subtask": "move grasped pot to left side of blue cloth"}, {"step": 9, "subtask": "release pot"}, {"step": 10, "subtask": "grasp pot"}, {"step": 11, "subtask": "move grasped pot to center of blue cloth"}, {"step": 12, "subtask": "release pot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/18cfac129f28f5ff3deb2ab2.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0496/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_rss_toykitchen6_pnp_sweep_68_train/10", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 496, "taskvar": "numpy_256_rss_toykitchen6_pnp_sweep_68_train", "episode_id": "10", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 451, "role": "error_localization"}}
{"state": {"task": "Move the can to the lower side of the right burner, then to the upper side, then back to the lower side.", "plan": [{"step": 1, "subtask": "grasp can"}, {"step": 2, "subtask": "release can"}, {"step": 3, "subtask": "move grasped can to the lower side of the right burner"}, {"step": 4, "subtask": "release can"}, {"step": 5, "subtask": "grasp can"}, {"step": 6, "subtask": "move grasped can to the upper side of the right burner"}, {"step": 7, "subtask": "release can"}, {"step": 8, "subtask": "grasp can"}, {"step": 9, "subtask": "move grasped can to the lower side of the right burner"}, {"step": 10, "subtask": "release can"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/0c8de888bf45a9546a6c42bb.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0497/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train/13", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 497, "taskvar": "numpy_256_icra_toykitchen_fixed_cam_resetfree_push_sweep_toykitchen6_05_train", "episode_id": "13", "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 416, "role": "error_localization"}}
{"state": {"task": "take the banana out of the drawer and put it on the right side of the table", "plan": [{"step": 1, "subtask": "open drawer"}, {"step": 2, "subtask": "pick up banana"}, {"step": 3, "subtask": "release banana"}, {"step": 4, "subtask": "move grasped banana to the right side of the table"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/9bb4d4184bab31fef9ebd10a.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0498/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_11_train/15", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 498, "taskvar": "numpy_256_bridge_data_v2_datacol2_tabletop_dark_wood_drawer_pnp_11_train", "episode_id": "15", "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 301, "role": "error_localization"}}
{"state": {"task": "put spoon in bowl, then take bowl out of sink onto the table, put spoon in cabinet, take another spoon from cabinet, put in bowl", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon to bowl"}, {"step": 3, "subtask": "release spoon"}, {"step": 4, "subtask": "grasp bowl"}, {"step": 5, "subtask": "move grasped bowl out"}, {"step": 6, "subtask": "move grasped bowl to the table"}, {"step": 7, "subtask": "release bowl"}, {"step": 8, "subtask": "grasp cabinet door handle"}, {"step": 9, "subtask": "move grasped door handle to open position"}, {"step": 10, "subtask": "release door handle"}, {"step": 11, "subtask": "grasp spoon"}, {"step": 12, "subtask": "move grasped spoon to cabinet"}, {"step": 13, "subtask": "release spoon"}, {"step": 14, "subtask": "grasp fork"}, {"step": 15, "subtask": "move grasped object to bowl"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a WidowX robot arm with a parallel gripper (BridgeData V2 toy kitchens and tabletops) before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "bdv2fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/bdv2fail/7b4dadefeb1fa68fbf04460e.png"}], "_meta": {"source": "bdv2fail", "origin": "paulpacaud/bdv2fail_test_dataset@3d84d34df6ca5a59a598651ef453fd68de09de06", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "bdv2fail/plan0499/bdv2fail_plan_failure_mode", "group_id": "bdv2fail/numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train/7", "family": "bdv2fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "BDV2-Fail (BridgeDataV2-Fail)", "native_line": 499, "taskvar": "numpy_256_bridge_data_v1_berkeley_toykitchen6_put_spoon_in_bowl_sink_train", "episode_id": "7", "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 511, "role": "error_localization"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "subtask": "Pick up apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/77f8cbc29be4c1738da1e4bb.png"}, {"type": "image", "uri": "media/robofail/36ad754f0d3984f905cd834a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0000/robofail_exec_success", "group_id": "robofail/putAppleBowl1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 0, "taskvar": "putAppleBowl1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 218, "role": "outcome"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "subtask": "Put apple inside bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/36ad754f0d3984f905cd834a.png"}, {"type": "image", "uri": "media/robofail/64a50624c36aa633c011e6a0.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0001/robofail_exec_success", "group_id": "robofail/putAppleBowl1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 1, "taskvar": "putAppleBowl1", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "subtask": "Pick up apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/fa3d0f06acacd2a31e5571e1.png"}, {"type": "image", "uri": "media/robofail/2d131179e8355d1ae7073589.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0002/robofail_exec_success", "group_id": "robofail/putAppleBowl2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 2, "taskvar": "putAppleBowl2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 218, "role": "outcome"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "subtask": "Put apple inside bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/2d131179e8355d1ae7073589.png"}, {"type": "image", "uri": "media/robofail/fa4369d2e99eb4a5ba37c6e3.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0003/robofail_exec_success", "group_id": "robofail/putAppleBowl2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 3, "taskvar": "putAppleBowl2", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "slip", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "subtask": "Pick up apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/620f44e69a2b80525e07574e.png"}, {"type": "image", "uri": "media/robofail/21bff2c37fb260c36a65ecd7.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0004/robofail_exec_success", "group_id": "robofail/putAppleBowl3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 4, "taskvar": "putAppleBowl3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 218, "role": "outcome"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "subtask": "Put apple inside bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/c280da0219e49622ac51a3fd.png"}, {"type": "image", "uri": "media/robofail/c4aeb27c7d03c449d0c27d6a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0005/robofail_exec_success", "group_id": "robofail/putAppleBowl3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 5, "taskvar": "putAppleBowl3", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 219, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/fdb9962fa3a4fc0792610807.png"}, {"type": "image", "uri": "media/robofail/3f1ea52482ef57a0d5ff0304.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0006/robofail_exec_success", "group_id": "robofail/boilWater1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 6, "taskvar": "boilWater1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Put pot in sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/3f1ea52482ef57a0d5ff0304.png"}, {"type": "image", "uri": "media/robofail/300a5ecaadafde7a484033a0.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0007/robofail_exec_success", "group_id": "robofail/boilWater1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 7, "taskvar": "boilWater1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle on faucet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/01f6e49c089a6db3f58829ba.png"}, {"type": "image", "uri": "media/robofail/87b48ae67e56297ea9dfab6b.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0008/robofail_exec_success", "group_id": "robofail/boilWater1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 8, "taskvar": "boilWater1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle off faucet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/2e34058bb4b483dace988b0d.png"}, {"type": "image", "uri": "media/robofail/f16d6872ea40fdbca0c041ea.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0009/robofail_exec_success", "group_id": "robofail/boilWater1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 9, "taskvar": "boilWater1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/f16d6872ea40fdbca0c041ea.png"}, {"type": "image", "uri": "media/robofail/667142e40a15ba99f90ef63d.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0010/robofail_exec_success", "group_id": "robofail/boilWater1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 10, "taskvar": "boilWater1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Put bowl on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/62d6ffd9d6ae8f23e4909a8a.png"}, {"type": "image", "uri": "media/robofail/1d926bb43b1574f6b678c4ce.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0011/robofail_exec_success", "group_id": "robofail/boilWater1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 11, "taskvar": "boilWater1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/1d926bb43b1574f6b678c4ce.png"}, {"type": "image", "uri": "media/robofail/8f84eabb7d0fc67198c08fbf.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0012/robofail_exec_success", "group_id": "robofail/boilWater1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 12, "taskvar": "boilWater1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/915c57d4dbe2dba052d046db.png"}, {"type": "image", "uri": "media/robofail/f3697f7b9a6f4e137328b98e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0013/robofail_exec_success", "group_id": "robofail/boilWater2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 13, "taskvar": "boilWater2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Put pot in sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/f3697f7b9a6f4e137328b98e.png"}, {"type": "image", "uri": "media/robofail/b7eab003a7aec61e8c536a85.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0014/robofail_exec_success", "group_id": "robofail/boilWater2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 14, "taskvar": "boilWater2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/f59e3dbb429861caa4cc0c11.png"}, {"type": "image", "uri": "media/robofail/5ab505e30876f0e248ad77af.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0015/robofail_exec_success", "group_id": "robofail/boilWater2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 15, "taskvar": "boilWater2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Put pot on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/5ab505e30876f0e248ad77af.png"}, {"type": "image", "uri": "media/robofail/5ae93a763ff622947126b124.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0016/robofail_exec_success", "group_id": "robofail/boilWater2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 16, "taskvar": "boilWater2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/5ae93a763ff622947126b124.png"}, {"type": "image", "uri": "media/robofail/1f78d823858353a6c516383a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0017/robofail_exec_success", "group_id": "robofail/boilWater2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 17, "taskvar": "boilWater2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/3fad12c6d59c536c30ed9e9a.png"}, {"type": "image", "uri": "media/robofail/5a15ab41e507e4753eb54d6e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0018/robofail_exec_success", "group_id": "robofail/boilWater3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 18, "taskvar": "boilWater3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Put pot in sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/86604e48b99e8feae8901570.png"}, {"type": "image", "uri": "media/robofail/53af21afa7303fff449b0a9c.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0019/robofail_exec_success", "group_id": "robofail/boilWater3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 19, "taskvar": "boilWater3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle on faucet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/53af21afa7303fff449b0a9c.png"}, {"type": "image", "uri": "media/robofail/ebe219e0f3b7b10f8072654b.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0020/robofail_exec_success", "group_id": "robofail/boilWater3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 20, "taskvar": "boilWater3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle off faucet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/f17f82b76099cacdc20f25dd.png"}, {"type": "image", "uri": "media/robofail/368412aa10f56a24dfdf1f05.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0021/robofail_exec_success", "group_id": "robofail/boilWater3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 21, "taskvar": "boilWater3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/3720d16a38fff59118768eef.png"}, {"type": "image", "uri": "media/robofail/fc2de736ff9493d11e5b4129.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0022/robofail_exec_success", "group_id": "robofail/boilWater3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 22, "taskvar": "boilWater3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Put pot on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/af531f02c8a9a70c2f7a6307.png"}, {"type": "image", "uri": "media/robofail/468176b4fc3c8aba2a92fc70.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0023/robofail_exec_success", "group_id": "robofail/boilWater3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 23, "taskvar": "boilWater3", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/d486c0174decd82c578340a3.png"}, {"type": "image", "uri": "media/robofail/f952b1e9fcb33bcdda740d62.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0024/robofail_exec_success", "group_id": "robofail/boilWater3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 24, "taskvar": "boilWater3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/7785c7e316f997d5027e9572.png"}, {"type": "image", "uri": "media/robofail/c5320f83976b3e2dc5cc0641.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0025/robofail_exec_success", "group_id": "robofail/boilWater4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 25, "taskvar": "boilWater4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Put pot in sink", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/2a04499a334d648eb046af85.png"}, {"type": "image", "uri": "media/robofail/9ce1f53996ab86ecf21fcb9c.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0026/robofail_exec_success", "group_id": "robofail/boilWater4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 26, "taskvar": "boilWater4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle on faucet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/8ae4794cef5b01478dbd1fc0.png"}, {"type": "image", "uri": "media/robofail/8759814487f400f0c79b10d4.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0027/robofail_exec_success", "group_id": "robofail/boilWater4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 27, "taskvar": "boilWater4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle off faucet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/1cc16c7fbe85f697749f1501.png"}, {"type": "image", "uri": "media/robofail/060b67f31710fee02baadaa6.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0028/robofail_exec_success", "group_id": "robofail/boilWater4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 28, "taskvar": "boilWater4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/4a41f09ed52dd0b00aea055d.png"}, {"type": "image", "uri": "media/robofail/0f6848bb0f1c366f7ac6935a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0029/robofail_exec_success", "group_id": "robofail/boilWater4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 29, "taskvar": "boilWater4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Put pot on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/0f6848bb0f1c366f7ac6935a.png"}, {"type": "image", "uri": "media/robofail/6cfc27b47136a8364fdcd3a2.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0030/robofail_exec_success", "group_id": "robofail/boilWater4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 30, "taskvar": "boilWater4", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "slip", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/6cfc27b47136a8364fdcd3a2.png"}, {"type": "image", "uri": "media/robofail/1f5f0fc6a724e408df9157b4.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0031/robofail_exec_success", "group_id": "robofail/boilWater4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 31, "taskvar": "boilWater4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "cut carrot, so that at the end, carrot is sliced", "subtask": "Pick up knife", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/6580f144ec74b7b8c7c45eb5.png"}, {"type": "image", "uri": "media/robofail/1d60a09061d65d032d87903e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0032/robofail_exec_success", "group_id": "robofail/cutCarrot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 32, "taskvar": "cutCarrot1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 215, "role": "outcome"}}
{"state": {"task": "cut carrot, so that at the end, carrot is sliced", "subtask": "Slice carrot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/1d60a09061d65d032d87903e.png"}, {"type": "image", "uri": "media/robofail/a824cadcbb3882f270282875.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0033/robofail_exec_success", "group_id": "robofail/cutCarrot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 33, "taskvar": "cutCarrot1", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 214, "role": "outcome"}}
{"state": {"task": "cut carrot, so that at the end, carrot is sliced", "subtask": "Pick up knife", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/5f90025ac22836894e24aa01.png"}, {"type": "image", "uri": "media/robofail/b4272500e730ce7584436cb7.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0034/robofail_exec_success", "group_id": "robofail/cutCarrot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 34, "taskvar": "cutCarrot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 215, "role": "outcome"}}
{"state": {"task": "cut carrot, so that at the end, carrot is sliced", "subtask": "Slice carrot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/1222164b829b086a9985c905.png"}, {"type": "image", "uri": "media/robofail/9249159e728114f93297d06f.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0035/robofail_exec_success", "group_id": "robofail/cutCarrot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 35, "taskvar": "cutCarrot2", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "slip", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 214, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Pick up pear", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/c0b3bf6e4159bb2d34b31759.png"}, {"type": "image", "uri": "media/robofail/4ab5d661718ab4db57b1c824.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0036/robofail_exec_success", "group_id": "robofail/putPearDrawer1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 36, "taskvar": "putPearDrawer1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Put pear in drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/4ab5d661718ab4db57b1c824.png"}, {"type": "image", "uri": "media/robofail/b2ca8406deb4902b9bd9c06e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0037/robofail_exec_success", "group_id": "robofail/putPearDrawer1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 37, "taskvar": "putPearDrawer1", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Open drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b5d30ee302583ef424c3694b.png"}, {"type": "image", "uri": "media/robofail/99c29edb36d03a6c3ed0e488.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0038/robofail_exec_success", "group_id": "robofail/putPearDrawer2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 38, "taskvar": "putPearDrawer2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Pick up pear", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/99c29edb36d03a6c3ed0e488.png"}, {"type": "image", "uri": "media/robofail/3620a42cb5f3414f5438f77b.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0039/robofail_exec_success", "group_id": "robofail/putPearDrawer2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 39, "taskvar": "putPearDrawer2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Put pear in drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/3620a42cb5f3414f5438f77b.png"}, {"type": "image", "uri": "media/robofail/f4f1cb3a95004726356d2f57.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0040/robofail_exec_success", "group_id": "robofail/putPearDrawer2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 40, "taskvar": "putPearDrawer2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Close drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/f4f1cb3a95004726356d2f57.png"}, {"type": "image", "uri": "media/robofail/c23a69074c01c9e4bea1be33.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0041/robofail_exec_success", "group_id": "robofail/putPearDrawer2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 41, "taskvar": "putPearDrawer2", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Pick up pear", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/a302beb14bd5fe9b22b7ff65.png"}, {"type": "image", "uri": "media/robofail/b27e430fc16c3f38460c5444.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0042/robofail_exec_success", "group_id": "robofail/putPearDrawer3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 42, "taskvar": "putPearDrawer3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Open drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b27e430fc16c3f38460c5444.png"}, {"type": "image", "uri": "media/robofail/5e0f597d078967e0d7a0b08f.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0043/robofail_exec_success", "group_id": "robofail/putPearDrawer3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 43, "taskvar": "putPearDrawer3", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Put pear in drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/81bfa32482fc32ecc5ed69da.png"}, {"type": "image", "uri": "media/robofail/d328fb4ec3684ded92240ac7.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0044/robofail_exec_success", "group_id": "robofail/putPearDrawer3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 44, "taskvar": "putPearDrawer3", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "subtask": "Close drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/d328fb4ec3684ded92240ac7.png"}, {"type": "image", "uri": "media/robofail/78a74cc1554754e9ee7b2118.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0045/robofail_exec_success", "group_id": "robofail/putPearDrawer3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 45, "taskvar": "putPearDrawer3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Pick up knife", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/55c7525bf843af4d37310679.png"}, {"type": "image", "uri": "media/robofail/932cf13c4d4d4e9e9a092184.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0046/robofail_exec_success", "group_id": "robofail/sauteeCarrot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 46, "taskvar": "sauteeCarrot1", "episode_id": null, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Slice carrot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/932cf13c4d4d4e9e9a092184.png"}, {"type": "image", "uri": "media/robofail/654b835307912eb7d837eb79.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0047/robofail_exec_success", "group_id": "robofail/sauteeCarrot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 47, "taskvar": "sauteeCarrot1", "episode_id": null, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Put knife on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/654b835307912eb7d837eb79.png"}, {"type": "image", "uri": "media/robofail/0e49a852d19742b337ca90ee.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0048/robofail_exec_success", "group_id": "robofail/sauteeCarrot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 48, "taskvar": "sauteeCarrot1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Pick up carrot slice", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/11c56494b7d6dce8b398eecd.png"}, {"type": "image", "uri": "media/robofail/2d0b2eebaec30aceaa5d3223.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0049/robofail_exec_success", "group_id": "robofail/sauteeCarrot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 49, "taskvar": "sauteeCarrot1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Put carrot slice in saucepan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/2d0b2eebaec30aceaa5d3223.png"}, {"type": "image", "uri": "media/robofail/47e6b3b8175c7b26c2ceefd5.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0050/robofail_exec_success", "group_id": "robofail/sauteeCarrot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 50, "taskvar": "sauteeCarrot1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/47e6b3b8175c7b26c2ceefd5.png"}, {"type": "image", "uri": "media/robofail/ada947a830d11b33948fa398.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0051/robofail_exec_success", "group_id": "robofail/sauteeCarrot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 51, "taskvar": "sauteeCarrot1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Pick up knife", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/ee21bb51adab193cb9f3bf86.png"}, {"type": "image", "uri": "media/robofail/e42cbc6da99b89bfe2e2c00c.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0052/robofail_exec_success", "group_id": "robofail/sauteeCarrot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 52, "taskvar": "sauteeCarrot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Slice carrot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b150ba21a244171ee66b43f1.png"}, {"type": "image", "uri": "media/robofail/68634b4b2a3e0e77a39f169b.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0053/robofail_exec_success", "group_id": "robofail/sauteeCarrot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 53, "taskvar": "sauteeCarrot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Put knife on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/1d85b8b574e51fa96d2fb163.png"}, {"type": "image", "uri": "media/robofail/9d5f6fdace7420c15e0710b9.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0054/robofail_exec_success", "group_id": "robofail/sauteeCarrot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 54, "taskvar": "sauteeCarrot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Pick up carrot slice", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b7dbaeaf2ed5db4c28777ae3.png"}, {"type": "image", "uri": "media/robofail/a1f97ee2d1867c86c4e1d844.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0055/robofail_exec_success", "group_id": "robofail/sauteeCarrot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 55, "taskvar": "sauteeCarrot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Put carrot slice in saucepan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/a1f97ee2d1867c86c4e1d844.png"}, {"type": "image", "uri": "media/robofail/668e2c11cc21f2113f5b75dd.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0056/robofail_exec_success", "group_id": "robofail/sauteeCarrot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 56, "taskvar": "sauteeCarrot2", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/668e2c11cc21f2113f5b75dd.png"}, {"type": "image", "uri": "media/robofail/0e07d3ea0bcf8ef722937e9c.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0057/robofail_exec_success", "group_id": "robofail/sauteeCarrot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 57, "taskvar": "sauteeCarrot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Pick up knife", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b4adc571b4f5e631d852695a.png"}, {"type": "image", "uri": "media/robofail/d983fa1429ac0b1c7111937f.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0058/robofail_exec_success", "group_id": "robofail/sauteeCarrot3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 58, "taskvar": "sauteeCarrot3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Slice carrot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/e999e2c5fe1d3e3866045a46.png"}, {"type": "image", "uri": "media/robofail/2bb5ab021e6abc035a7225d8.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0059/robofail_exec_success", "group_id": "robofail/sauteeCarrot3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 59, "taskvar": "sauteeCarrot3", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "slip", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Put knife on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/7e61efe3b508bba58d93aba0.png"}, {"type": "image", "uri": "media/robofail/ca80a5a0fc96b8aae31054e7.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0060/robofail_exec_success", "group_id": "robofail/sauteeCarrot3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 60, "taskvar": "sauteeCarrot3", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "slip", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Pick up carrot slice", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/ca80a5a0fc96b8aae31054e7.png"}, {"type": "image", "uri": "media/robofail/358239e480de580b260b7a8e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0061/robofail_exec_success", "group_id": "robofail/sauteeCarrot3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 61, "taskvar": "sauteeCarrot3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Put carrot slice in saucepan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/358239e480de580b260b7a8e.png"}, {"type": "image", "uri": "media/robofail/4a238d5bcdc87ec764e8fbec.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0062/robofail_exec_success", "group_id": "robofail/sauteeCarrot3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 62, "taskvar": "sauteeCarrot3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/402885bed81598b2a2a94486.png"}, {"type": "image", "uri": "media/robofail/5d787eca40f4853ae82a0272.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0063/robofail_exec_success", "group_id": "robofail/sauteeCarrot3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 63, "taskvar": "sauteeCarrot3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Pick up knife", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/d28a10816dfe25c57aee67de.png"}, {"type": "image", "uri": "media/robofail/a3321e793c42bbe4b390ecb4.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0064/robofail_exec_success", "group_id": "robofail/sauteeCarrot4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 64, "taskvar": "sauteeCarrot4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Slice carrot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/9bffe2f60134ad50239803f9.png"}, {"type": "image", "uri": "media/robofail/9527e2f4a96cfeefbaf64427.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0065/robofail_exec_success", "group_id": "robofail/sauteeCarrot4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 65, "taskvar": "sauteeCarrot4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Put knife on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/9527e2f4a96cfeefbaf64427.png"}, {"type": "image", "uri": "media/robofail/e6ba63f123b138d4f8bf3534.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0066/robofail_exec_success", "group_id": "robofail/sauteeCarrot4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 66, "taskvar": "sauteeCarrot4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Pick up carrot slice", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/fb4be7fbb6021fd36435aae1.png"}, {"type": "image", "uri": "media/robofail/1f1c15158bec9e46608e48b9.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0067/robofail_exec_success", "group_id": "robofail/sauteeCarrot4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 67, "taskvar": "sauteeCarrot4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "subtask": "Put carrot slice in saucepan", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/3a90f140382d06caf58e2c9e.png"}, {"type": "image", "uri": "media/robofail/a47ee9e4a9387896a326aa40.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0068/robofail_exec_success", "group_id": "robofail/sauteeCarrot4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 68, "taskvar": "sauteeCarrot4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Pick up knife", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/ad8caa2c2178abee953ffa26.png"}, {"type": "image", "uri": "media/robofail/2860ab588d4b77c36facc0d4.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0069/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 69, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Put knife in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/2860ab588d4b77c36facc0d4.png"}, {"type": "image", "uri": "media/robofail/df94f7c8c1a8eb3293b82d91.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0070/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 70, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 224, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Close fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/df94f7c8c1a8eb3293b82d91.png"}, {"type": "image", "uri": "media/robofail/91dd0ebfeef86dcca60f3b3d.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0071/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 71, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Pick up pear", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/9af588f3ae377ee845ae7474.png"}, {"type": "image", "uri": "media/robofail/2741c91fad0ef8b80e92c7f0.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0072/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 72, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Put pear near drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/2741c91fad0ef8b80e92c7f0.png"}, {"type": "image", "uri": "media/robofail/ae4fa090267fc9c954b4e593.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0073/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 73, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Open drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/ae4fa090267fc9c954b4e593.png"}, {"type": "image", "uri": "media/robofail/44e9881af29c1c001d979aae.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0074/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 74, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Pick up pear", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/44e9881af29c1c001d979aae.png"}, {"type": "image", "uri": "media/robofail/858e8a82079e02dea279ca8d.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0075/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 75, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Put pear in the fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/4d70188ac980b3047fbb5963.png"}, {"type": "image", "uri": "media/robofail/3a79e6fad2ed0a0ddd0465f0.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0076/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 76, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 224, "role": "outcome"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "subtask": "Close drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/a4b93a959eba7d0d3f853c5f.png"}, {"type": "image", "uri": "media/robofail/f6f1b35ddba6197fd05dd42f.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0077/robofail_exec_success", "group_id": "robofail/secureObjects1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 77, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put all visible fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Pick up strawberry", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/5d23b4c5b7b89cd220e6f25e.png"}, {"type": "image", "uri": "media/robofail/5b1920d518bb1e0e96e08c8e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0078/robofail_exec_success", "group_id": "robofail/putFruitsBowl1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 78, "taskvar": "putFruitsBowl1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "put all visible fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Put strawberry in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/520dc6a02c38b3d3ad515016.png"}, {"type": "image", "uri": "media/robofail/3689a8bbfc6e711cb7439e0c.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0079/robofail_exec_success", "group_id": "robofail/putFruitsBowl1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 79, "taskvar": "putFruitsBowl1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put all visible fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Pick up pear", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/65271946510541857dfa4c10.png"}, {"type": "image", "uri": "media/robofail/e12a46785fbbc5a113f33d92.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0080/robofail_exec_success", "group_id": "robofail/putFruitsBowl1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 80, "taskvar": "putFruitsBowl1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "put all visible fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Put pear in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/e7a2830382bf2b55cc808886.png"}, {"type": "image", "uri": "media/robofail/9ba16c5eba70a34b4a763963.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0081/robofail_exec_success", "group_id": "robofail/putFruitsBowl1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 81, "taskvar": "putFruitsBowl1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put all visible fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Pick up carrot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/9ba16c5eba70a34b4a763963.png"}, {"type": "image", "uri": "media/robofail/40455d8bd9b166ec97241230.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0082/robofail_exec_success", "group_id": "robofail/putFruitsBowl1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 82, "taskvar": "putFruitsBowl1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "put all visible fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Put carrot in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b2d5393a9d22d949d04b1ad4.png"}, {"type": "image", "uri": "media/robofail/6bd48a057c9693fcdc1d4512.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0083/robofail_exec_success", "group_id": "robofail/putFruitsBowl1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 83, "taskvar": "putFruitsBowl1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 223, "role": "outcome"}}
{"state": {"task": "put all fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Pick up apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/6d6a9080f631403f08f678f8.png"}, {"type": "image", "uri": "media/robofail/cc4786122d76f1b47bd9f6a5.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0084/robofail_exec_success", "group_id": "robofail/putFruitsBowl2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 84, "taskvar": "putFruitsBowl2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put all fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Put apple in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/cc4786122d76f1b47bd9f6a5.png"}, {"type": "image", "uri": "media/robofail/ab8a2be1154b229e3c60c7f1.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0085/robofail_exec_success", "group_id": "robofail/putFruitsBowl2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 85, "taskvar": "putFruitsBowl2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "put all fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Pick up pear", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/fde53b54650a99dc07beeb3f.png"}, {"type": "image", "uri": "media/robofail/d74b975831a417434dd2b464.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0086/robofail_exec_success", "group_id": "robofail/putFruitsBowl2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 86, "taskvar": "putFruitsBowl2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "put all fruits in bowl, so that at the end, all fruits are in the bowl", "subtask": "Put pear in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/188886020368c3ea91aec7d2.png"}, {"type": "image", "uri": "media/robofail/c015edbe60953c1e0d3bb5b3.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0087/robofail_exec_success", "group_id": "robofail/putFruitsBowl2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 87, "taskvar": "putFruitsBowl2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Pick up cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/13576547988e45aaaa856285.png"}, {"type": "image", "uri": "media/robofail/adef7d4c0fd318424e201cf9.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0088/robofail_exec_success", "group_id": "robofail/makeCoffee1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 88, "taskvar": "makeCoffee1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 220, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Put cup in coffee machine", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/40a3aa75fd20a5e8c193cba3.png"}, {"type": "image", "uri": "media/robofail/8c6390a0ba26edc9986a5bda.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0089/robofail_exec_success", "group_id": "robofail/makeCoffee1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 89, "taskvar": "makeCoffee1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Toggle on coffee machine", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/8c6390a0ba26edc9986a5bda.png"}, {"type": "image", "uri": "media/robofail/2ec1859a3bba4840f5d17fd1.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0090/robofail_exec_success", "group_id": "robofail/makeCoffee1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 90, "taskvar": "makeCoffee1", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Pick up cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/2ec1859a3bba4840f5d17fd1.png"}, {"type": "image", "uri": "media/robofail/0b0acd109a7c85d8cea5d2f9.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0091/robofail_exec_success", "group_id": "robofail/makeCoffee1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 91, "taskvar": "makeCoffee1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 220, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Put cup on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/0b0acd109a7c85d8cea5d2f9.png"}, {"type": "image", "uri": "media/robofail/b3732b8dbc34378974bb7791.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0092/robofail_exec_success", "group_id": "robofail/makeCoffee1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 92, "taskvar": "makeCoffee1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Pick up cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/f597e16a30845ce6fc6aea23.png"}, {"type": "image", "uri": "media/robofail/0cb76116ee4f0ea1a1354a06.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0093/robofail_exec_success", "group_id": "robofail/makeCoffee2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 93, "taskvar": "makeCoffee2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 220, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Put cup in coffee machine", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/0cb76116ee4f0ea1a1354a06.png"}, {"type": "image", "uri": "media/robofail/ec7a30f0daa6a58a72661787.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0094/robofail_exec_success", "group_id": "robofail/makeCoffee2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 94, "taskvar": "makeCoffee2", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Toggle on coffee machine", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/1aa89b4e62429b1648ab703a.png"}, {"type": "image", "uri": "media/robofail/0149fced8d23b8f38298c9ed.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0095/robofail_exec_success", "group_id": "robofail/makeCoffee2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 95, "taskvar": "makeCoffee2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Pick up cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/0149fced8d23b8f38298c9ed.png"}, {"type": "image", "uri": "media/robofail/0da154316efcac0c2e29ac4f.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0096/robofail_exec_success", "group_id": "robofail/makeCoffee2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 96, "taskvar": "makeCoffee2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 220, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "subtask": "Put cup on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/0da154316efcac0c2e29ac4f.png"}, {"type": "image", "uri": "media/robofail/e0312865aa4a9ca69958e56b.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0097/robofail_exec_success", "group_id": "robofail/makeCoffee2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 97, "taskvar": "makeCoffee2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a blue cup filled with coffee is on table", "subtask": "Toggle on coffee machine", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/a72e25b5d76d1d2a97ed085f.png"}, {"type": "image", "uri": "media/robofail/b494a4a15f1c0c6c67c2e6ab.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0098/robofail_exec_success", "group_id": "robofail/makeCoffee3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 98, "taskvar": "makeCoffee3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a blue cup filled with coffee is on table", "subtask": "Pick up cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b494a4a15f1c0c6c67c2e6ab.png"}, {"type": "image", "uri": "media/robofail/b7d576b98e2ab0d0aad05528.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0099/robofail_exec_success", "group_id": "robofail/makeCoffee3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 99, "taskvar": "makeCoffee3", "episode_id": null, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 221, "role": "outcome"}}
{"state": {"task": "make coffee, so that at the end, a blue cup filled with coffee is on table", "subtask": "Put cup on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/413008c830c8a16fe5753a7b.png"}, {"type": "image", "uri": "media/robofail/be8245ace1b9e7921e2630a6.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0100/robofail_exec_success", "group_id": "robofail/makeCoffee3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 100, "taskvar": "makeCoffee3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 222, "role": "outcome"}}
{"state": {"task": "pre-heat pot, so that at the end, a pot is on top of a stove burner that is toggled on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/54f7e3f57264fba79076a4ef.png"}, {"type": "image", "uri": "media/robofail/458e42cfecfb1a028b2fd58b.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0101/robofail_exec_success", "group_id": "robofail/heatPot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 101, "taskvar": "heatPot1", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "imprecise grasping/pushing", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "pre-heat pot, so that at the end, a pot is on top of a stove burner that is toggled on", "subtask": "Put pot on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/458e42cfecfb1a028b2fd58b.png"}, {"type": "image", "uri": "media/robofail/eb7afbccab86f0a30641924a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0102/robofail_exec_success", "group_id": "robofail/heatPot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 102, "taskvar": "heatPot1", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pre-heat pot, so that at the end, a pot is on top of a stove burner that is toggled on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/eb7afbccab86f0a30641924a.png"}, {"type": "image", "uri": "media/robofail/d1f799df88ab53ed963f5f0c.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0103/robofail_exec_success", "group_id": "robofail/heatPot1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 103, "taskvar": "heatPot1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "pre-heat pot, so that at the end, a pot is on top of a stove burner that is toggled on", "subtask": "Pick up pot", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/35a8e7f605aa88ddae109145.png"}, {"type": "image", "uri": "media/robofail/8aae15ca9b5284fad8be4fa9.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0104/robofail_exec_success", "group_id": "robofail/heatPot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 104, "taskvar": "heatPot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "pre-heat pot, so that at the end, a pot is on top of a stove burner that is toggled on", "subtask": "Put pot on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/8aae15ca9b5284fad8be4fa9.png"}, {"type": "image", "uri": "media/robofail/af5666d308cf91cb7603b99a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0105/robofail_exec_success", "group_id": "robofail/heatPot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 105, "taskvar": "heatPot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pre-heat pot, so that at the end, a pot is on top of a stove burner that is toggled on", "subtask": "Toggle on stove burner", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/af5666d308cf91cb7603b99a.png"}, {"type": "image", "uri": "media/robofail/510b7d161e362429ab0afe45.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0106/robofail_exec_success", "group_id": "robofail/heatPot2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 106, "taskvar": "heatPot2", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Pick up apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/c2aa353b1f0c9b28625f65ab.png"}, {"type": "image", "uri": "media/robofail/95e5b1ef59a8ceca95b11491.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0107/robofail_exec_success", "group_id": "robofail/appleInFridge1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 107, "taskvar": "appleInFridge1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Put apple in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/8d40199f34d8a009ba577cfc.png"}, {"type": "image", "uri": "media/robofail/d07dee14ea74651ec158702e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0108/robofail_exec_success", "group_id": "robofail/appleInFridge1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 108, "taskvar": "appleInFridge1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Open fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/aabea2f8f4490de231e04d35.png"}, {"type": "image", "uri": "media/robofail/038c9fe841f3dba4800a4abe.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0109/robofail_exec_success", "group_id": "robofail/appleInFridge1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 109, "taskvar": "appleInFridge1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/038c9fe841f3dba4800a4abe.png"}, {"type": "image", "uri": "media/robofail/e38715ed0258fef0b4aec782.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0110/robofail_exec_success", "group_id": "robofail/appleInFridge1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 110, "taskvar": "appleInFridge1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Put bowl in fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/e38715ed0258fef0b4aec782.png"}, {"type": "image", "uri": "media/robofail/990782113abc13c8a8a4fc96.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0111/robofail_exec_success", "group_id": "robofail/appleInFridge1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 111, "taskvar": "appleInFridge1", "episode_id": null, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Close fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/990782113abc13c8a8a4fc96.png"}, {"type": "image", "uri": "media/robofail/27dc5a134f4fe96466714e86.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0112/robofail_exec_success", "group_id": "robofail/appleInFridge1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 112, "taskvar": "appleInFridge1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Pick up apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/ac3c1d65eee0fb56270cbbf0.png"}, {"type": "image", "uri": "media/robofail/2fbf95990a8c726c4e4d77d9.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0113/robofail_exec_success", "group_id": "robofail/appleInFridge2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 113, "taskvar": "appleInFridge2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Put apple in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/945cc4047e849926192269fe.png"}, {"type": "image", "uri": "media/robofail/880be10be21c494276110fd0.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0114/robofail_exec_success", "group_id": "robofail/appleInFridge2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 114, "taskvar": "appleInFridge2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Open fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/4efba5efb22a470d8f4b7115.png"}, {"type": "image", "uri": "media/robofail/6f8b482a28f7fca4ac712988.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0115/robofail_exec_success", "group_id": "robofail/appleInFridge2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 115, "taskvar": "appleInFridge2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/9275a788ca77bfc2cea6baad.png"}, {"type": "image", "uri": "media/robofail/158f310daf70c54fe27007bf.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0116/robofail_exec_success", "group_id": "robofail/appleInFridge2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 116, "taskvar": "appleInFridge2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Put bowl in fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/158f310daf70c54fe27007bf.png"}, {"type": "image", "uri": "media/robofail/dc45f6f44e2f664625eea479.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0117/robofail_exec_success", "group_id": "robofail/appleInFridge2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 117, "taskvar": "appleInFridge2", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Close fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/abf22c6734cbbfac468e2b37.png"}, {"type": "image", "uri": "media/robofail/03b8a7dff604d6991e6d7ca3.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0118/robofail_exec_success", "group_id": "robofail/appleInFridge2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 118, "taskvar": "appleInFridge2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Pick up apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/fe3c5521c7d97f5854e1fa00.png"}, {"type": "image", "uri": "media/robofail/8e0163e0727f3a377e741008.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0119/robofail_exec_success", "group_id": "robofail/appleInFridge3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 119, "taskvar": "appleInFridge3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Put apple in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/8e0163e0727f3a377e741008.png"}, {"type": "image", "uri": "media/robofail/d8d00729cbd1e13b8f7779c3.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0120/robofail_exec_success", "group_id": "robofail/appleInFridge3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 120, "taskvar": "appleInFridge3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/d8d00729cbd1e13b8f7779c3.png"}, {"type": "image", "uri": "media/robofail/291ab8716fdc0e97d8912128.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0121/robofail_exec_success", "group_id": "robofail/appleInFridge3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 121, "taskvar": "appleInFridge3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Put bowl in fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/291ab8716fdc0e97d8912128.png"}, {"type": "image", "uri": "media/robofail/e793b9fbeaf10fad9f3659b8.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0122/robofail_exec_success", "group_id": "robofail/appleInFridge3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 122, "taskvar": "appleInFridge3", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Open fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/e793b9fbeaf10fad9f3659b8.png"}, {"type": "image", "uri": "media/robofail/eed2c89f1e10d89de9833471.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0123/robofail_exec_success", "group_id": "robofail/appleInFridge3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 123, "taskvar": "appleInFridge3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "subtask": "Close fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/eed2c89f1e10d89de9833471.png"}, {"type": "image", "uri": "media/robofail/9a4bb88ba1c901d9dfefc14e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0124/robofail_exec_success", "group_id": "robofail/appleInFridge3", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 124, "taskvar": "appleInFridge3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in a closed fridge", "subtask": "Pick up apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/d9ee2163a56a6fae9fbf42e8.png"}, {"type": "image", "uri": "media/robofail/b0890a2c27855f4ff273e95b.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0125/robofail_exec_success", "group_id": "robofail/appleInFridge4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 125, "taskvar": "appleInFridge4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in a closed fridge", "subtask": "Put apple in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/70e92019d885682608495ddf.png"}, {"type": "image", "uri": "media/robofail/3ca9d8416a6c2fe4271252aa.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0126/robofail_exec_success", "group_id": "robofail/appleInFridge4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 126, "taskvar": "appleInFridge4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in a closed fridge", "subtask": "Open fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/3ca9d8416a6c2fe4271252aa.png"}, {"type": "image", "uri": "media/robofail/097c2b5a70d75f994ded1336.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0127/robofail_exec_success", "group_id": "robofail/appleInFridge4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 127, "taskvar": "appleInFridge4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in a closed fridge", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/097c2b5a70d75f994ded1336.png"}, {"type": "image", "uri": "media/robofail/0f1d4e335d349e6aa18ba6db.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0128/robofail_exec_success", "group_id": "robofail/appleInFridge4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 128, "taskvar": "appleInFridge4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in a closed fridge", "subtask": "Put bowl in fridge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/ccaa7e8b8f95a0393151aa90.png"}, {"type": "image", "uri": "media/robofail/b2f9a310ae8bceb4076dc7f1.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0129/robofail_exec_success", "group_id": "robofail/appleInFridge4", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 129, "taskvar": "appleInFridge4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Pick up potato", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/9bab132938ffdf5c08acc439.png"}, {"type": "image", "uri": "media/robofail/6bd4703d1f4dfdb00f5fef15.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0130/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 130, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Put potato in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/6bd4703d1f4dfdb00f5fef15.png"}, {"type": "image", "uri": "media/robofail/0aaf21d73e26b474a204e645.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0131/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 131, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Open microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/812174535a25b6f78096815f.png"}, {"type": "image", "uri": "media/robofail/aa6afb6d89cbad9c368328c6.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0132/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 132, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/aa6afb6d89cbad9c368328c6.png"}, {"type": "image", "uri": "media/robofail/daff49f8671fc83925f43a7d.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0133/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 133, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Put bowl in microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/daff49f8671fc83925f43a7d.png"}, {"type": "image", "uri": "media/robofail/b77601ac33a74317b2d19e03.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0134/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 134, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Toggle on microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b77601ac33a74317b2d19e03.png"}, {"type": "image", "uri": "media/robofail/4b9a06f04cf098e018b8ee63.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0135/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 135, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Close microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/4b9a06f04cf098e018b8ee63.png"}, {"type": "image", "uri": "media/robofail/41c6f2fe4f4862c05a64f750.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0136/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 136, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Toggle off microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/41c6f2fe4f4862c05a64f750.png"}, {"type": "image", "uri": "media/robofail/69155086bd62afd007246c0e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0137/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 137, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Open microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/b81764c0864a9284a0bec546.png"}, {"type": "image", "uri": "media/robofail/349be6396d81d976b14b55a1.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0138/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 138, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/9da9b722c23bc35161dd0d4d.png"}, {"type": "image", "uri": "media/robofail/7f09db6b72a2448101d15f65.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0139/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 139, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Put bowl on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/7f09db6b72a2448101d15f65.png"}, {"type": "image", "uri": "media/robofail/d867575ac51ccdd6fea2606f.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0140/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 140, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Close microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/d867575ac51ccdd6fea2606f.png"}, {"type": "image", "uri": "media/robofail/0cdf05a38ab5479f0644eb23.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0141/robofail_exec_success", "group_id": "robofail/heatPotato1", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 141, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Pick up potato", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/caf0bcc0488acecb04a1b7ea.png"}, {"type": "image", "uri": "media/robofail/ea31c553473e73aafc9dd808.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0142/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 142, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Put potato in bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/26aa05b933f79e5640eeb50b.png"}, {"type": "image", "uri": "media/robofail/294b9189d12ea840ac1a4ed7.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0143/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 143, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Toggle on microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/294b9189d12ea840ac1a4ed7.png"}, {"type": "image", "uri": "media/robofail/e92ab0cb311c0c59b663696c.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0144/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 144, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Open microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/d9d5ce9a473ec8962121f88c.png"}, {"type": "image", "uri": "media/robofail/a0ab75aa9256f2ce82b9e33f.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0145/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 145, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "imprecise grasping/pushing", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Open microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/a23924f9395fdb6fa63e4e6b.png"}, {"type": "image", "uri": "media/robofail/81f3f41789c54f2333df3a70.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0146/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 146, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/81f3f41789c54f2333df3a70.png"}, {"type": "image", "uri": "media/robofail/16230470bf4127341772fbb8.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0147/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 147, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Put bowl in microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/102be1878e8f93a42fb2ea17.png"}, {"type": "image", "uri": "media/robofail/39513305f8dc5b41d3bb2d13.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0148/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 148, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Close microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/39513305f8dc5b41d3bb2d13.png"}, {"type": "image", "uri": "media/robofail/7808f47e35b6d571a664432a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0149/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 149, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Open microwave", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/00672f64165edd4b7d5ca50e.png"}, {"type": "image", "uri": "media/robofail/de6c6511d9a28cbbf57bcdc4.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0150/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 150, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Pick up bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/de6c6511d9a28cbbf57bcdc4.png"}, {"type": "image", "uri": "media/robofail/a0468624805ed7d0c961c103.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0151/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 151, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "subtask": "Put bowl on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot arm with a gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robofail_exec_success"}}, "media": [{"type": "image", "uri": "media/robofail/a0468624805ed7d0c961c103.png"}, {"type": "image", "uri": "media/robofail/a2a41faf064a30c878103e56.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/exec0152/robofail_exec_success", "group_id": "robofail/heatPotato2", "family": "robofail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 152, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720], [1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "plan": [{"step": 1, "subtask": "Pick up apple"}, {"step": 2, "subtask": "Put apple inside bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/77f8cbc29be4c1738da1e4bb.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0000/robofail_plan_correct", "group_id": "robofail/putAppleBowl1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 0, "taskvar": "putAppleBowl1", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "plan": [{"step": 1, "subtask": "Pick up apple"}, {"step": 2, "subtask": "Put apple inside bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/fa3d0f06acacd2a31e5571e1.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0001/robofail_plan_correct", "group_id": "robofail/putAppleBowl2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 1, "taskvar": "putAppleBowl2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "put apple in bowl, so that at the end, apple is inside bowl", "plan": [{"step": 1, "subtask": "Pick up apple"}, {"step": 2, "subtask": "Put apple inside bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/620f44e69a2b80525e07574e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0002/robofail_plan_correct", "group_id": "robofail/putAppleBowl3", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 2, "taskvar": "putAppleBowl3", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 221, "role": "error_localization"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "plan": [{"step": 1, "subtask": "Pick up pot"}, {"step": 2, "subtask": "Put pot in sink"}, {"step": 3, "subtask": "Toggle on faucet"}, {"step": 4, "subtask": "Toggle off faucet"}, {"step": 5, "subtask": "Pick up bowl"}, {"step": 6, "subtask": "Put bowl on stove burner"}, {"step": 7, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/fdb9962fa3a4fc0792610807.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0003/robofail_plan_correct", "group_id": "robofail/boilWater1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 3, "taskvar": "boilWater1", "episode_id": null, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 317, "role": "error_localization"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "plan": [{"step": 1, "subtask": "Pick up pot"}, {"step": 2, "subtask": "Put pot in sink"}, {"step": 3, "subtask": "Pick up pot"}, {"step": 4, "subtask": "Put pot on stove burner"}, {"step": 5, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/915c57d4dbe2dba052d046db.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0004/robofail_plan_correct", "group_id": "robofail/boilWater2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 4, "taskvar": "boilWater2", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 287, "role": "error_localization"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "plan": [{"step": 1, "subtask": "Pick up pot"}, {"step": 2, "subtask": "Put pot in sink"}, {"step": 3, "subtask": "Toggle on faucet"}, {"step": 4, "subtask": "Toggle off faucet"}, {"step": 5, "subtask": "Pick up pot"}, {"step": 6, "subtask": "Put pot on stove burner"}, {"step": 7, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/3fad12c6d59c536c30ed9e9a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0005/robofail_plan_correct", "group_id": "robofail/boilWater3", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 5, "taskvar": "boilWater3", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 317, "role": "error_localization"}}
{"state": {"task": "boil water in a pot, so that at the end, a pot is filled with water, the pot is on top of a stove burner that is turned on", "plan": [{"step": 1, "subtask": "Pick up pot"}, {"step": 2, "subtask": "Put pot in sink"}, {"step": 3, "subtask": "Toggle on faucet"}, {"step": 4, "subtask": "Toggle off faucet"}, {"step": 5, "subtask": "Pick up pot"}, {"step": 6, "subtask": "Put pot on stove burner"}, {"step": 7, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/7785c7e316f997d5027e9572.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0006/robofail_plan_correct", "group_id": "robofail/boilWater4", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 6, "taskvar": "boilWater4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 317, "role": "error_localization"}}
{"state": {"task": "cut carrot, so that at the end, carrot is sliced", "plan": [{"step": 1, "subtask": "Pick up knife"}, {"step": 2, "subtask": "Slice carrot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/6580f144ec74b7b8c7c45eb5.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0007/robofail_plan_correct", "group_id": "robofail/cutCarrot1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 7, "taskvar": "cutCarrot1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 216, "role": "error_localization"}}
{"state": {"task": "cut carrot, so that at the end, carrot is sliced", "plan": [{"step": 1, "subtask": "Pick up knife"}, {"step": 2, "subtask": "Slice carrot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/5f90025ac22836894e24aa01.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0008/robofail_plan_correct", "group_id": "robofail/cutCarrot2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 8, "taskvar": "cutCarrot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 216, "role": "error_localization"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "plan": [{"step": 1, "subtask": "Pick up pear"}, {"step": 2, "subtask": "Put pear in drawer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/c0b3bf6e4159bb2d34b31759.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0009/robofail_plan_correct", "group_id": "robofail/putPearDrawer1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 9, "taskvar": "putPearDrawer1", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 225, "role": "error_localization"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "plan": [{"step": 1, "subtask": "Open drawer"}, {"step": 2, "subtask": "Pick up pear"}, {"step": 3, "subtask": "Put pear in drawer"}, {"step": 4, "subtask": "Close drawer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/b5d30ee302583ef424c3694b.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0010/robofail_plan_correct", "group_id": "robofail/putPearDrawer2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 10, "taskvar": "putPearDrawer2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 253, "role": "error_localization"}}
{"state": {"task": "put pear in drawer, so that at the end, pear is inside drawer and drawer is closed", "plan": [{"step": 1, "subtask": "Pick up pear"}, {"step": 2, "subtask": "Open drawer"}, {"step": 3, "subtask": "Put pear in drawer"}, {"step": 4, "subtask": "Close drawer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/a302beb14bd5fe9b22b7ff65.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0011/robofail_plan_correct", "group_id": "robofail/putPearDrawer3", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 11, "taskvar": "putPearDrawer3", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 253, "role": "error_localization"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "plan": [{"step": 1, "subtask": "Pick up knife"}, {"step": 2, "subtask": "Slice carrot"}, {"step": 3, "subtask": "Put knife on table"}, {"step": 4, "subtask": "Pick up carrot slice"}, {"step": 5, "subtask": "Put carrot slice in saucepan"}, {"step": 6, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/55c7525bf843af4d37310679.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0012/robofail_plan_correct", "group_id": "robofail/sauteeCarrot1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 12, "taskvar": "sauteeCarrot1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 300, "role": "error_localization"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "plan": [{"step": 1, "subtask": "Pick up knife"}, {"step": 2, "subtask": "Slice carrot"}, {"step": 3, "subtask": "Put knife on table"}, {"step": 4, "subtask": "Pick up carrot slice"}, {"step": 5, "subtask": "Put carrot slice in saucepan"}, {"step": 6, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/ee21bb51adab193cb9f3bf86.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0013/robofail_plan_correct", "group_id": "robofail/sauteeCarrot2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 13, "taskvar": "sauteeCarrot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 300, "role": "error_localization"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "plan": [{"step": 1, "subtask": "Pick up knife"}, {"step": 2, "subtask": "Slice carrot"}, {"step": 3, "subtask": "Put knife on table"}, {"step": 4, "subtask": "Pick up carrot slice"}, {"step": 5, "subtask": "Put carrot slice in saucepan"}, {"step": 6, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/b4adc571b4f5e631d852695a.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0014/robofail_plan_correct", "group_id": "robofail/sauteeCarrot3", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 14, "taskvar": "sauteeCarrot3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 300, "role": "error_localization"}}
{"state": {"task": "cook carrot, so that at the end, a sliced carrot is inside a pan, the pan is on top of a stove burner that is turned on", "plan": [{"step": 1, "subtask": "Pick up knife"}, {"step": 2, "subtask": "Slice carrot"}, {"step": 3, "subtask": "Put knife on table"}, {"step": 4, "subtask": "Pick up carrot slice"}, {"step": 5, "subtask": "Put carrot slice in saucepan"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/d28a10816dfe25c57aee67de.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0015/robofail_plan_correct", "group_id": "robofail/sauteeCarrot4", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 15, "taskvar": "sauteeCarrot4", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 284, "role": "error_localization"}}
{"state": {"task": "secure objects, so that at the end, pear and knife are in their most suitable container respectively", "plan": [{"step": 1, "subtask": "Pick up knife"}, {"step": 2, "subtask": "Put knife in the drawer"}, {"step": 3, "subtask": "Close fridge"}, {"step": 4, "subtask": "Pick up pear"}, {"step": 5, "subtask": "Put pear near drawer"}, {"step": 6, "subtask": "Open drawer"}, {"step": 7, "subtask": "Pick up pear"}, {"step": 8, "subtask": "Put pear in the fridge"}, {"step": 9, "subtask": "Close drawer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/ad8caa2c2178abee953ffa26.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0016/robofail_plan_correct", "group_id": "robofail/secureObjects1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 16, "taskvar": "secureObjects1", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 331, "role": "error_localization"}}
{"state": {"task": "put all visible fruits in bowl, so that at the end, all fruits are in the bowl", "plan": [{"step": 1, "subtask": "Pick up strawberry"}, {"step": 2, "subtask": "Put strawberry in bowl"}, {"step": 3, "subtask": "Pick up pear"}, {"step": 4, "subtask": "Put pear in bowl"}, {"step": 5, "subtask": "Pick up carrot"}, {"step": 6, "subtask": "Put carrot in bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/5d23b4c5b7b89cd220e6f25e.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0017/robofail_plan_correct", "group_id": "robofail/putFruitsBowl1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 17, "taskvar": "putFruitsBowl1", "episode_id": null, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 287, "role": "error_localization"}}
{"state": {"task": "put all fruits in bowl, so that at the end, all fruits are in the bowl", "plan": [{"step": 1, "subtask": "Pick up apple"}, {"step": 2, "subtask": "Put apple in bowl"}, {"step": 3, "subtask": "Pick up pear"}, {"step": 4, "subtask": "Put pear in bowl"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/6d6a9080f631403f08f678f8.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0018/robofail_plan_correct", "group_id": "robofail/putFruitsBowl2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 18, "taskvar": "putFruitsBowl2", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 255, "role": "error_localization"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "plan": [{"step": 1, "subtask": "Pick up cup"}, {"step": 2, "subtask": "Put cup in coffee machine"}, {"step": 3, "subtask": "Toggle on coffee machine"}, {"step": 4, "subtask": "Pick up cup"}, {"step": 5, "subtask": "Put cup on table"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/13576547988e45aaaa856285.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0019/robofail_plan_correct", "group_id": "robofail/makeCoffee1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 19, "taskvar": "makeCoffee1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 271, "role": "error_localization"}}
{"state": {"task": "make coffee, so that at the end, a cup filled with coffee is on table", "plan": [{"step": 1, "subtask": "Pick up cup"}, {"step": 2, "subtask": "Put cup in coffee machine"}, {"step": 3, "subtask": "Toggle on coffee machine"}, {"step": 4, "subtask": "Pick up cup"}, {"step": 5, "subtask": "Put cup on table"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/f597e16a30845ce6fc6aea23.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0020/robofail_plan_correct", "group_id": "robofail/makeCoffee2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 20, "taskvar": "makeCoffee2", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 271, "role": "error_localization"}}
{"state": {"task": "make coffee, so that at the end, a blue cup filled with coffee is on table", "plan": [{"step": 1, "subtask": "Toggle on coffee machine"}, {"step": 2, "subtask": "Pick up cup"}, {"step": 3, "subtask": "Put cup on table"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/a72e25b5d76d1d2a97ed085f.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0021/robofail_plan_correct", "group_id": "robofail/makeCoffee3", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 21, "taskvar": "makeCoffee3", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 240, "role": "error_localization"}}
{"state": {"task": "pre-heat pot, so that at the end, a pot is on top of a stove burner that is toggled on", "plan": [{"step": 1, "subtask": "Pick up pot"}, {"step": 2, "subtask": "Put pot on stove burner"}, {"step": 3, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/54f7e3f57264fba79076a4ef.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0022/robofail_plan_correct", "group_id": "robofail/heatPot1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 22, "taskvar": "heatPot1", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 248, "role": "error_localization"}}
{"state": {"task": "pre-heat pot, so that at the end, a pot is on top of a stove burner that is toggled on", "plan": [{"step": 1, "subtask": "Pick up pot"}, {"step": 2, "subtask": "Put pot on stove burner"}, {"step": 3, "subtask": "Toggle on stove burner"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/35a8e7f605aa88ddae109145.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0023/robofail_plan_correct", "group_id": "robofail/heatPot2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 23, "taskvar": "heatPot2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 248, "role": "error_localization"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "plan": [{"step": 1, "subtask": "Pick up apple"}, {"step": 2, "subtask": "Put apple in bowl"}, {"step": 3, "subtask": "Open fridge"}, {"step": 4, "subtask": "Pick up bowl"}, {"step": 5, "subtask": "Put bowl in fridge"}, {"step": 6, "subtask": "Close fridge"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/c2aa353b1f0c9b28625f65ab.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0024/robofail_plan_correct", "group_id": "robofail/appleInFridge1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 24, "taskvar": "appleInFridge1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 292, "role": "error_localization"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "plan": [{"step": 1, "subtask": "Pick up apple"}, {"step": 2, "subtask": "Put apple in bowl"}, {"step": 3, "subtask": "Open fridge"}, {"step": 4, "subtask": "Pick up bowl"}, {"step": 5, "subtask": "Put bowl in fridge"}, {"step": 6, "subtask": "Close fridge"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/ac3c1d65eee0fb56270cbbf0.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0025/robofail_plan_correct", "group_id": "robofail/appleInFridge2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 25, "taskvar": "appleInFridge2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 292, "role": "error_localization"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in fridge", "plan": [{"step": 1, "subtask": "Pick up apple"}, {"step": 2, "subtask": "Put apple in bowl"}, {"step": 3, "subtask": "Pick up bowl"}, {"step": 4, "subtask": "Put bowl in fridge"}, {"step": 5, "subtask": "Open fridge"}, {"step": 6, "subtask": "Close fridge"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/fe3c5521c7d97f5854e1fa00.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0026/robofail_plan_correct", "group_id": "robofail/appleInFridge3", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 26, "taskvar": "appleInFridge3", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 292, "role": "error_localization"}}
{"state": {"task": "store apple in a bowl and put in a fridge, so that at the end, a bowl with an apple inside is put in a closed fridge", "plan": [{"step": 1, "subtask": "Pick up apple"}, {"step": 2, "subtask": "Put apple in bowl"}, {"step": 3, "subtask": "Open fridge"}, {"step": 4, "subtask": "Pick up bowl"}, {"step": 5, "subtask": "Put bowl in fridge"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/d9ee2163a56a6fae9fbf42e8.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0027/robofail_plan_correct", "group_id": "robofail/appleInFridge4", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 27, "taskvar": "appleInFridge4", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 280, "role": "error_localization"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "plan": [{"step": 1, "subtask": "Pick up potato"}, {"step": 2, "subtask": "Put potato in bowl"}, {"step": 3, "subtask": "Open microwave"}, {"step": 4, "subtask": "Pick up bowl"}, {"step": 5, "subtask": "Put bowl in microwave"}, {"step": 6, "subtask": "Toggle on microwave"}, {"step": 7, "subtask": "Close microwave"}, {"step": 8, "subtask": "Toggle off microwave"}, {"step": 9, "subtask": "Open microwave"}, {"step": 10, "subtask": "Pick up bowl"}, {"step": 11, "subtask": "Put bowl on table"}, {"step": 12, "subtask": "Close microwave"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/9bab132938ffdf5c08acc439.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0028/robofail_plan_correct", "group_id": "robofail/heatPotato1", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 28, "taskvar": "heatPotato1", "episode_id": null, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "serve a bowl of potato on table that was heated using a microwave, so that at the end, a bowl of a heated potato is on the table", "plan": [{"step": 1, "subtask": "Pick up potato"}, {"step": 2, "subtask": "Put potato in bowl"}, {"step": 3, "subtask": "Toggle on microwave"}, {"step": 4, "subtask": "Open microwave"}, {"step": 5, "subtask": "Open microwave"}, {"step": 6, "subtask": "Pick up bowl"}, {"step": 7, "subtask": "Put bowl in microwave"}, {"step": 8, "subtask": "Close microwave"}, {"step": 9, "subtask": "Open microwave"}, {"step": 10, "subtask": "Pick up bowl"}, {"step": 11, "subtask": "Put bowl on table"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a robot arm with a gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "robofail_plan_correct"}}, "media": [{"type": "image", "uri": "media/robofail/caf0bcc0488acecb04a1b7ea.png"}], "_meta": {"source": "robofail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robofail/plan0029/robofail_plan_correct", "group_id": "robofail/heatPotato2", "family": "robofail_plan_correct", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboFail", "native_line": 29, "taskvar": "heatPotato2", "episode_id": null, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[1280, 720]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a900eed38d04f158b37c1ad3.mp4"}, {"type": "video", "uri": "media/robometer_ood/715f5ef324357037d32ec950.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000002-vs-video_000030/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["33480409-b84b-4b36-8e2b-f04918cc1819", "faf4bcb1-d384-403c-829f-53b965fcd58d"], "files": ["mit_franka/video_000002.mp4", "mit_franka/video_000030.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5789c1b330d1fc030fab68ad.mp4"}, {"type": "video", "uri": "media/robometer_ood/279067a6ee71b3bd6541c6f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000003-vs-video_000025/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["163574f2-46c2-486c-9a18-e7a8265bef61", "000c9dd6-e04c-4ee5-8af4-3abf3314c673"], "files": ["mit_franka/video_000003.mp4", "mit_franka/video_000025.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2eefd5cc8b26a719a181ce06.mp4"}, {"type": "video", "uri": "media/robometer_ood/22d772e51515af4e88a7df8a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000004-vs-video_000028/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["a23e5584-65c1-4895-91bd-2cbf2f7a3f58", "c0806df4-7289-4c26-be73-40b4a33b471b"], "files": ["mit_franka/video_000004.mp4", "mit_franka/video_000028.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a3b994fe7be669f93bbf4bbe.mp4"}, {"type": "video", "uri": "media/robometer_ood/d0b1bcf4aea85db2a8c8f872.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000008-vs-video_000029/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["01e3be9a-96b0-45b4-a15d-9827a50ba098", "efb6b2d7-60e0-4b26-82cd-3e07464521d5"], "files": ["mit_franka/video_000029.mp4", "mit_franka/video_000008.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/91feff49c974a4f8f2529294.mp4"}, {"type": "video", "uri": "media/robometer_ood/27c4f48937bfd2b505635e0f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000011-vs-video_000031/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["3cd81bbe-28bd-4149-8ecd-1c54425b4f45", "8c333dd9-3d99-448e-ae58-b681b36d502b"], "files": ["mit_franka/video_000011.mp4", "mit_franka/video_000031.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/077caac87189f78905bef638.mp4"}, {"type": "video", "uri": "media/robometer_ood/10d2670f4c5d94a15f755a69.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000016-vs-video_000024/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["cb618f9a-2475-4bd0-8f03-aab64e31cb1a", "9004622a-b59a-4aab-b02c-6d0ac4c4bf79"], "files": ["mit_franka/video_000016.mp4", "mit_franka/video_000024.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cbc8127ed56bc4cf45c2f8fd.mp4"}, {"type": "video", "uri": "media/robometer_ood/6a2bf7c4415884e3849b4492.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000017-vs-video_000027/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["14a26a41-f949-4c2a-aced-4d5d0b9f6f6f", "8d7ec266-6bb4-4ba0-91ea-3241abdf90a6"], "files": ["mit_franka/video_000027.mp4", "mit_franka/video_000017.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1b00e8d6604d7fa73693df94.mp4"}, {"type": "video", "uri": "media/robometer_ood/cbcd9b0abcc30f4d9b720fa3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000020-vs-video_000026/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["7a6d915b-9383-4c99-ac43-30b3e6ddb340", "03ade13b-792a-451b-9ea8-94485f758d07"], "files": ["mit_franka/video_000026.mp4", "mit_franka/video_000020.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5789c1b330d1fc030fab68ad.mp4"}, {"type": "video", "uri": "media/robometer_ood/75e35dba7d2bc3f3a640c2dd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000032-vs-video_000003/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["163574f2-46c2-486c-9a18-e7a8265bef61", "45e39a64-fc77-4650-844d-8a2d82dc934d"], "files": ["mit_franka/video_000003.mp4", "mit_franka/video_000032.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a900eed38d04f158b37c1ad3.mp4"}, {"type": "video", "uri": "media/robometer_ood/938ae535abbbe185a71632bf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000033-vs-video_000002/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["33480409-b84b-4b36-8e2b-f04918cc1819", "741f8ab9-6d38-457e-95c9-4e52a1605535"], "files": ["mit_franka/video_000002.mp4", "mit_franka/video_000033.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cbc8127ed56bc4cf45c2f8fd.mp4"}, {"type": "video", "uri": "media/robometer_ood/938ae535abbbe185a71632bf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000033-vs-video_000027/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["14a26a41-f949-4c2a-aced-4d5d0b9f6f6f", "741f8ab9-6d38-457e-95c9-4e52a1605535"], "files": ["mit_franka/video_000027.mp4", "mit_franka/video_000033.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/21af52420afe700175d004f7.mp4"}, {"type": "video", "uri": "media/robometer_ood/c14e604a12a1b0b7980bb45f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000034-vs-video_000022/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["3d5c6e20-011d-4f47-b4df-082bbf0d1509", "bf994e7d-3f4a-40c4-9da8-cd464d98f373"], "files": ["mit_franka/video_000022.mp4", "mit_franka/video_000034.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b2991ab6bbd57a978f8fa4dc.mp4"}, {"type": "video", "uri": "media/robometer_ood/655fcba18d192a8f256e994d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000035-vs-video_000021/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["6b423129-f607-4ae2-8e84-4cf98ba5b7fa", "868ddcdd-3ca3-49a2-ad96-e74059926a14"], "files": ["mit_franka/video_000035.mp4", "mit_franka/video_000021.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/10d2670f4c5d94a15f755a69.mp4"}, {"type": "video", "uri": "media/robometer_ood/b2991ab6bbd57a978f8fa4dc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000035-vs-video_000024/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["9004622a-b59a-4aab-b02c-6d0ac4c4bf79", "6b423129-f607-4ae2-8e84-4cf98ba5b7fa"], "files": ["mit_franka/video_000024.mp4", "mit_franka/video_000035.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3de5c58d4f2cb4c474951184.mp4"}, {"type": "video", "uri": "media/robometer_ood/39aab8098cee5194e9d33e02.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000036-vs-video_000023/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["ec5cc6e4-1e92-4d88-8a88-9b82a40f51ce", "0bc5e5cb-61af-4adc-8586-a193f76e780b"], "files": ["mit_franka/video_000036.mp4", "mit_franka/video_000023.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/98d852169ae8002741a10791.mp4"}, {"type": "video", "uri": "media/robometer_ood/78676aff96bb3d6aa4ab2d78.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000037-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["bf3b8897-238d-4a87-8bad-3dfc94318d1c", "05538a11-c296-466f-8865-e4e3c0ea5d58"], "files": ["mit_franka/video_000001.mp4", "mit_franka/video_000037.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0950432353cf90624731a736.mp4"}, {"type": "video", "uri": "media/robometer_ood/de22cd784e97579a05cd9b7d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000038-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["07091ac9-b6b3-4a05-bf89-f414182ed8a8", "042a6a74-794a-4b19-a664-212617d4dae3"], "files": ["mit_franka/video_000038.mp4", "mit_franka/video_000007.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0950432353cf90624731a736.mp4"}, {"type": "video", "uri": "media/robometer_ood/715f5ef324357037d32ec950.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000038-vs-video_000030/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["07091ac9-b6b3-4a05-bf89-f414182ed8a8", "faf4bcb1-d384-403c-829f-53b965fcd58d"], "files": ["mit_franka/video_000038.mp4", "mit_franka/video_000030.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9eec637eaee3d3a44ab4ef21.mp4"}, {"type": "video", "uri": "media/robometer_ood/65d45d1162d7ea79bd1a2859.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000039-vs-video_000010/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["4e7942d0-1367-4a96-9a94-361dbb9515b0", "dd34d3ce-3758-4543-8bd1-c10135b46e71"], "files": ["mit_franka/video_000010.mp4", "mit_franka/video_000039.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4158a20718beed24f74eec54.mp4"}, {"type": "video", "uri": "media/robometer_ood/b71667f656ada7742b705293.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000040-vs-video_000009/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["60928009-148a-4a91-bd41-837b7554fbea", "8beeb483-c9bd-4e6b-a2a7-fba7557db07e"], "files": ["mit_franka/video_000009.mp4", "mit_franka/video_000040.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f6273f7bc3891f91027a836c.mp4"}, {"type": "video", "uri": "media/robometer_ood/b4aacbdd9888a547c8c8b39a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000041-vs-video_000012/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["619b454a-d5ab-4b6e-ba20-2dad1e55b3f6", "49163975-14fa-4df9-b022-361889a59960"], "files": ["mit_franka/video_000012.mp4", "mit_franka/video_000041.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4cc8b569b9a1f65971e6825f.mp4"}, {"type": "video", "uri": "media/robometer_ood/bc1deaf37938f7173970bc23.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000042-vs-video_000015/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["7603a1bc-a20a-412c-92fa-76e4958ad58f", "d45cab77-4326-4111-bca6-575f19a2bb17"], "files": ["mit_franka/video_000042.mp4", "mit_franka/video_000015.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4cc8b569b9a1f65971e6825f.mp4"}, {"type": "video", "uri": "media/robometer_ood/279067a6ee71b3bd6541c6f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000042-vs-video_000025/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["7603a1bc-a20a-412c-92fa-76e4958ad58f", "000c9dd6-e04c-4ee5-8af4-3abf3314c673"], "files": ["mit_franka/video_000042.mp4", "mit_franka/video_000025.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b259c7110007e6b27d122d42.mp4"}, {"type": "video", "uri": "media/robometer_ood/252a166454efffc64a7a822b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000043-vs-video_000005/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["519d5f5d-7b81-4b7f-9134-d14af6922ab3", "ae8c2b1b-bda9-42ab-86bd-791e2db8e5ab"], "files": ["mit_franka/video_000043.mp4", "mit_franka/video_000005.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/447a18a11a34b07bc80bcb5a.mp4"}, {"type": "video", "uri": "media/robometer_ood/2777ee313d5bd4cbd9e93735.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000044-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["7336bc38-c543-4205-8882-829a2d4f4280", "ef82095d-6647-438b-ae41-33b46b745009"], "files": ["mit_franka/video_000006.mp4", "mit_franka/video_000044.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/077caac87189f78905bef638.mp4"}, {"type": "video", "uri": "media/robometer_ood/5bc5f0483e6a93e5b5709cfe.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000045-vs-video_000016/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["cb618f9a-2475-4bd0-8f03-aab64e31cb1a", "815000fd-8761-4399-a7e3-c57e20f82bc5"], "files": ["mit_franka/video_000016.mp4", "mit_franka/video_000045.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bcf2d84744fa69263e33af9a.mp4"}, {"type": "video", "uri": "media/robometer_ood/81aba2e82baae6ff2d10738f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000046-vs-video_000014/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["c77fedd0-526f-40a8-9a00-fd24fb4f450c", "b8554066-7766-43ac-a5de-5ac395f857b3"], "files": ["mit_franka/video_000046.mp4", "mit_franka/video_000014.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/17cb21b7b907e84744bdcc2f.mp4"}, {"type": "video", "uri": "media/robometer_ood/2eefd5cc8b26a719a181ce06.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000047-vs-video_000004/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["b8821982-69c0-4dcd-bdeb-ddce09d47295", "a23e5584-65c1-4895-91bd-2cbf2f7a3f58"], "files": ["mit_franka/video_000047.mp4", "mit_franka/video_000004.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/27c4f48937bfd2b505635e0f.mp4"}, {"type": "video", "uri": "media/robometer_ood/17cb21b7b907e84744bdcc2f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000047-vs-video_000031/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["8c333dd9-3d99-448e-ae58-b681b36d502b", "b8821982-69c0-4dcd-bdeb-ddce09d47295"], "files": ["mit_franka/video_000031.mp4", "mit_franka/video_000047.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5b88762414ab07b7d55f5e1f.mp4"}, {"type": "video", "uri": "media/robometer_ood/6a2bf7c4415884e3849b4492.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000048-vs-video_000017/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["9f5b84f6-26f9-40b1-8bb4-a133beb51837", "8d7ec266-6bb4-4ba0-91ea-3241abdf90a6"], "files": ["mit_franka/video_000048.mp4", "mit_franka/video_000017.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ad135ee3cfa289ef1f6a8b67.mp4"}, {"type": "video", "uri": "media/robometer_ood/2eb3d50557585a709f11bf1e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000049-vs-video_000013/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["914335e2-47c0-4c8c-9a27-59cff97f9807", "93c4705e-fbb4-4206-b247-e00e1adb56ed"], "files": ["mit_franka/video_000049.mp4", "mit_franka/video_000013.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4e785d76f0c123c39c4d4d13.mp4"}, {"type": "video", "uri": "media/robometer_ood/e738e4f040e4ed36463a1035.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000050-vs-video_000018/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["e55a6b32-2e81-47fd-bf93-7d455f587f8d", "e045022e-10eb-49a5-b06f-7d71c618ffb2"], "files": ["mit_franka/video_000050.mp4", "mit_franka/video_000018.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fc6f74b59c608850b1930e37.mp4"}, {"type": "video", "uri": "media/robometer_ood/d0b1bcf4aea85db2a8c8f872.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000051-vs-video_000008/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["8d12f346-fc65-40a8-8ffe-131ef8a43d2c", "efb6b2d7-60e0-4b26-82cd-3e07464521d5"], "files": ["mit_franka/video_000051.mp4", "mit_franka/video_000008.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fc6f74b59c608850b1930e37.mp4"}, {"type": "video", "uri": "media/robometer_ood/1b00e8d6604d7fa73693df94.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000051-vs-video_000026/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["8d12f346-fc65-40a8-8ffe-131ef8a43d2c", "7a6d915b-9383-4c99-ac43-30b3e6ddb340"], "files": ["mit_franka/video_000051.mp4", "mit_franka/video_000026.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/756a076957e189410aa89967.mp4"}, {"type": "video", "uri": "media/robometer_ood/1fe5f3eeb83147302274e686.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000052-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["0acaec09-1742-40c2-9751-b966d02d384c", "c7cfb350-f106-49f6-a3b5-ab0799faaf08"], "files": ["mit_franka/video_000019.mp4", "mit_franka/video_000052.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5c4ee05c9611a23cbe3cf3b2.mp4"}, {"type": "video", "uri": "media/robometer_ood/8e25ef44cda2d5c69c6124e9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000053-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["ab7117c2-9da2-4764-88d2-594da72cf144", "d12139fb-0bea-46b1-abfd-0dc23e000fa5"], "files": ["mit_franka/video_000000.mp4", "mit_franka/video_000053.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a3b994fe7be669f93bbf4bbe.mp4"}, {"type": "video", "uri": "media/robometer_ood/8e25ef44cda2d5c69c6124e9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000053-vs-video_000029/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["01e3be9a-96b0-45b4-a15d-9827a50ba098", "d12139fb-0bea-46b1-abfd-0dc23e000fa5"], "files": ["mit_franka/video_000029.mp4", "mit_franka/video_000053.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cbcd9b0abcc30f4d9b720fa3.mp4"}, {"type": "video", "uri": "media/robometer_ood/0e052d4e5ceab0c77630395e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000054-vs-video_000020/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["03ade13b-792a-451b-9ea8-94485f758d07", "6b5f3b7d-4869-455a-a689-432dceeca830"], "files": ["mit_franka/video_000020.mp4", "mit_franka/video_000054.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0e052d4e5ceab0c77630395e.mp4"}, {"type": "video", "uri": "media/robometer_ood/22d772e51515af4e88a7df8a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000054-vs-video_000028/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["6b5f3b7d-4869-455a-a689-432dceeca830", "c0806df4-7289-4c26-be73-40b4a33b471b"], "files": ["mit_franka/video_000054.mp4", "mit_franka/video_000028.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f7df15d61332db89c351712a.mp4"}, {"type": "video", "uri": "media/robometer_ood/91feff49c974a4f8f2529294.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000057-vs-video_000011/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["12ae16fd-d749-4a57-b6c3-9b8d4f08749d", "3cd81bbe-28bd-4149-8ecd-1c54425b4f45"], "files": ["mit_franka/video_000057.mp4", "mit_franka/video_000011.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/33b18a16f1791d24c2859ddd.mp4"}, {"type": "video", "uri": "media/robometer_ood/ce467e9743283e54135036b5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000058-vs-video_000094/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["6a963be3-befc-454d-907a-d6634557e682", "fd2789d8-d799-4fdf-9d09-a9c4ff3355ed"], "files": ["mit_franka/video_000094.mp4", "mit_franka/video_000058.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9b2becb7df94f5ea088a3db6.mp4"}, {"type": "video", "uri": "media/robometer_ood/0c6ceecef563b31cf2939f91.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000061-vs-video_000097/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["98de4790-789e-4eb0-870a-808ead842fef", "2f3fb3c1-4d80-4737-8da3-6113c4566ab6"], "files": ["mit_franka/video_000061.mp4", "mit_franka/video_000097.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6fd1eb026a7026e95190cbf9.mp4"}, {"type": "video", "uri": "media/robometer_ood/28f1c638b099f312b8c9c1b6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000062-vs-video_000087/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["b0a316a8-9cad-428f-a3e6-0662ff0412dd", "35a3dc57-1f65-435c-bf3b-193e8a104dd3"], "files": ["mit_franka/video_000087.mp4", "mit_franka/video_000062.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1daa40e150a06f3fe9fa1498.mp4"}, {"type": "video", "uri": "media/robometer_ood/3a4614d920eddb84437a942c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000063-vs-video_000093/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["2ebe5d8d-a81f-4626-95de-1941c78f3a23", "448a6d56-0a52-470b-942f-0744eb00ce5a"], "files": ["mit_franka/video_000093.mp4", "mit_franka/video_000063.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4c430e1d7bdffdf796cbd8f2.mp4"}, {"type": "video", "uri": "media/robometer_ood/c1e08aa7ceb16cbe244e48bd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000065-vs-video_000086/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["2e13e1b9-6f96-4bd2-bc7b-164be5bd329f", "20b51c66-336b-42df-a619-da72b8501547"], "files": ["mit_franka/video_000065.mp4", "mit_franka/video_000086.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f402cebaac8d6c8e69893b47.mp4"}, {"type": "video", "uri": "media/robometer_ood/4558f2e2282e8459e8fcde5e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000068-vs-video_000105/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["52b30f87-5541-4d97-976c-f1083c06764d", "7fb935c4-ac08-44a6-b46a-d0a406e9a128"], "files": ["mit_franka/video_000105.mp4", "mit_franka/video_000068.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/164a0f2143f62b2e06660ed3.mp4"}, {"type": "video", "uri": "media/robometer_ood/8fccc30dba7b991c22412112.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000069-vs-video_000104/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["944bc803-2cc3-459c-b340-cbdb6b54833c", "42965aef-8c98-4cc9-a49c-d7f8a3a6d86e"], "files": ["mit_franka/video_000069.mp4", "mit_franka/video_000104.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e4c60f6cd5360db7bdd8be6d.mp4"}, {"type": "video", "uri": "media/robometer_ood/f667ff391a6dc25fd532fac5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000070-vs-video_000091/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["6e7815c5-87da-4ec1-b6ee-6990c431ab96", "eb51e61f-1866-4fb9-814a-12b2f5a44298"], "files": ["mit_franka/video_000070.mp4", "mit_franka/video_000091.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/51df25f0dc82b86f12ea658a.mp4"}, {"type": "video", "uri": "media/robometer_ood/bdab1421cbe10716baab4501.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000072-vs-video_000092/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["f5b509e0-b261-4452-83e4-62683008d0f4", "f77fb1a0-c2a8-42da-9c9b-fcf6c04e5943"], "files": ["mit_franka/video_000092.mp4", "mit_franka/video_000072.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/44988f4bdfd0ad1b87ff18e5.mp4"}, {"type": "video", "uri": "media/robometer_ood/fa50f8ca47ce609f0ae567f9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000077-vs-video_000096/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["fd03f7c8-1fbb-4095-9fd0-58ddf8ab10f7", "3677a3d6-b3b4-485e-a126-f6bb39b2ddbc"], "files": ["mit_franka/video_000077.mp4", "mit_franka/video_000096.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fc5bafe3320b30dd6fd89aff.mp4"}, {"type": "video", "uri": "media/robometer_ood/0c4df98d982334e81b275a46.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000078-vs-video_000095/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["5a7d93e2-296f-4dd0-a804-4d11a5c2f0bf", "01baeffb-0ef4-4921-994f-e1c52a902d09"], "files": ["mit_franka/video_000078.mp4", "mit_franka/video_000095.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3ddf31acacfc6e6da329cfa9.mp4"}, {"type": "video", "uri": "media/robometer_ood/60688e494a8dcea966ffad72.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000079-vs-video_000099/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["b5dd5264-67f9-41c0-8aa6-07a103b19e23", "a71800dc-c84a-401c-b703-f38d6cb32389"], "files": ["mit_franka/video_000099.mp4", "mit_franka/video_000079.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b9f3708f551b5030e763f909.mp4"}, {"type": "video", "uri": "media/robometer_ood/ae5dd49bf572535a5de91011.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000082-vs-video_000089/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["07bd8aa3-9565-4931-a1ae-2b57622a89eb", "0c6ae2b2-273d-403a-ab87-d6c5e5550f99"], "files": ["mit_franka/video_000089.mp4", "mit_franka/video_000082.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/39e69f102052110a55dbd2dd.mp4"}, {"type": "video", "uri": "media/robometer_ood/32f9ebfc6076e809ba0809f3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000083-vs-video_000101/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["061f7ba2-297c-4ad3-aac5-a49e9aae9a16", "7e4c9f98-39c1-4984-b86b-4f60d9eec5a5"], "files": ["mit_franka/video_000083.mp4", "mit_franka/video_000101.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/85efd00b00b26190ea099729.mp4"}, {"type": "video", "uri": "media/robometer_ood/dd1c2e3e88559e36d3654382.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000084-vs-video_000098/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["8711a72c-b077-487d-82f5-eb271a00e839", "e751b226-a035-458a-a39d-d55c0c1c361c"], "files": ["mit_franka/video_000098.mp4", "mit_franka/video_000084.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2633672ecc4feb34e1c68628.mp4"}, {"type": "video", "uri": "media/robometer_ood/79ea9d39608f8f87e556ad57.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000088-vs-video_000100/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["83ccff35-571c-4a66-bf31-9d42097f5e1a", "800f0991-83b9-477a-b883-2b7abf401534"], "files": ["mit_franka/video_000100.mp4", "mit_franka/video_000088.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/dc9907017c1d48a2b36e36dc.mp4"}, {"type": "video", "uri": "media/robometer_ood/1876782203f1efab0706abcf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000103-vs-video_000074/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["e8dd3ee9-6d2b-4c0e-8dfb-732aad51c2ec", "e25bd65b-1441-4803-8bd0-266526d4396b"], "files": ["mit_franka/video_000074.mp4", "mit_franka/video_000103.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7af101e6cae6d9b79078761a.mp4"}, {"type": "video", "uri": "media/robometer_ood/9ee6c7551656e29f26692223.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000106-vs-video_000076/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["1de13c1d-4b72-437f-a1ed-4050f85f6557", "77d28dc7-bbca-4cb2-800a-2d41f1c243fc"], "files": ["mit_franka/video_000106.mp4", "mit_franka/video_000076.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bdab1421cbe10716baab4501.mp4"}, {"type": "video", "uri": "media/robometer_ood/af9785883c3dfc131a48ab10.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000107-vs-video_000072/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["f77fb1a0-c2a8-42da-9c9b-fcf6c04e5943", "eeb8b278-c25d-43bb-9df5-1813d46886cc"], "files": ["mit_franka/video_000072.mp4", "mit_franka/video_000107.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b9f3708f551b5030e763f909.mp4"}, {"type": "video", "uri": "media/robometer_ood/b2972d784d8515def4304a96.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000108-vs-video_000089/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["07bd8aa3-9565-4931-a1ae-2b57622a89eb", "1975e479-961e-459e-be6f-5eae24900626"], "files": ["mit_franka/video_000089.mp4", "mit_franka/video_000108.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b233aa87e62f212d6aae0763.mp4"}, {"type": "video", "uri": "media/robometer_ood/0f20629aa2f574fae9b3d387.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000112-vs-video_000066/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["0ee0335f-aa12-44dd-acd1-8344c231604b", "49d7bcef-33f6-46ce-bf99-d1f202eb8a2c"], "files": ["mit_franka/video_000112.mp4", "mit_franka/video_000066.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0b2b3d7bb0a54c265e1b068c.mp4"}, {"type": "video", "uri": "media/robometer_ood/7486f21678e7d69d1d055835.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000113-vs-video_000055/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["ea78fe7e-bebc-4a3a-bf02-5727074ff74e", "d92c5b5a-f552-4d0b-be41-091a5525149f"], "files": ["mit_franka/video_000055.mp4", "mit_franka/video_000113.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e9d9dbb8264d770a410fc3e6.mp4"}, {"type": "video", "uri": "media/robometer_ood/946b8bb8c3d1403c723b9c78.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000114-vs-video_000060/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["13ea62c3-4120-4fb1-9a91-fefe84b4fa03", "30bffa61-10ef-4870-bc86-24ec5e05f2ed"], "files": ["mit_franka/video_000060.mp4", "mit_franka/video_000114.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1a9379b9e580e83d91009253.mp4"}, {"type": "video", "uri": "media/robometer_ood/d3d02ad68a051f6d9595291d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000116-vs-video_000080/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["ee48378d-de05-4cb7-8df7-2715645e3d77", "4a81a138-b2ea-43e1-8a1c-1fc7497dbe3f"], "files": ["mit_franka/video_000080.mp4", "mit_franka/video_000116.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/51df25f0dc82b86f12ea658a.mp4"}, {"type": "video", "uri": "media/robometer_ood/d3d02ad68a051f6d9595291d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000116-vs-video_000092/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["f5b509e0-b261-4452-83e4-62683008d0f4", "4a81a138-b2ea-43e1-8a1c-1fc7497dbe3f"], "files": ["mit_franka/video_000092.mp4", "mit_franka/video_000116.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/baf65f48af730f6e7c91da96.mp4"}, {"type": "video", "uri": "media/robometer_ood/3a4614d920eddb84437a942c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000118-vs-video_000063/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["cf62479b-dec8-4b56-903d-9c66cf82e972", "448a6d56-0a52-470b-942f-0744eb00ce5a"], "files": ["mit_franka/video_000118.mp4", "mit_franka/video_000063.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/dd1c2e3e88559e36d3654382.mp4"}, {"type": "video", "uri": "media/robometer_ood/a0f3beed83982d4cfcddfa7b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000119-vs-video_000084/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["e751b226-a035-458a-a39d-d55c0c1c361c", "ede2ccec-63de-4297-9e03-c51508b83211"], "files": ["mit_franka/video_000084.mp4", "mit_franka/video_000119.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/42260c9bf042b8c1562c5181.mp4"}, {"type": "video", "uri": "media/robometer_ood/2633672ecc4feb34e1c68628.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000120-vs-video_000100/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["663b531c-43f0-4f96-af80-083bd06f827b", "83ccff35-571c-4a66-bf31-9d42097f5e1a"], "files": ["mit_franka/video_000120.mp4", "mit_franka/video_000100.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0d91ee883d87680b2ed31400.mp4"}, {"type": "video", "uri": "media/robometer_ood/adb8369b56b0403ae5ee4698.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000121-vs-video_000067/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["6c38f3c6-1258-4943-86bc-a84168201703", "d9967f26-14e0-4eec-a02b-982308054484"], "files": ["mit_franka/video_000121.mp4", "mit_franka/video_000067.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e4c60f6cd5360db7bdd8be6d.mp4"}, {"type": "video", "uri": "media/robometer_ood/7a893e3a67176a69aa4b05f4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000124-vs-video_000070/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["6e7815c5-87da-4ec1-b6ee-6990c431ab96", "15b63612-8916-4023-8f13-f7fb29add5df"], "files": ["mit_franka/video_000070.mp4", "mit_franka/video_000124.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bb123dfaed0703fb74363d9c.mp4"}, {"type": "video", "uri": "media/robometer_ood/a23579c4528c81bfe8300c72.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000126-vs-video_000073/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["9961c289-ba40-429b-8d89-1147d31dc746", "4343f624-7f22-4502-b84d-e608eac7f35b"], "files": ["mit_franka/video_000126.mp4", "mit_franka/video_000073.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/28f1c638b099f312b8c9c1b6.mp4"}, {"type": "video", "uri": "media/robometer_ood/d0f4862f7fa19f985556a805.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000129-vs-video_000062/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["35a3dc57-1f65-435c-bf3b-193e8a104dd3", "8cb48ffc-25d4-4337-a07d-963b24e888bc"], "files": ["mit_franka/video_000062.mp4", "mit_franka/video_000129.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cfdc87037bb3c1481af5b99c.mp4"}, {"type": "video", "uri": "media/robometer_ood/4c430e1d7bdffdf796cbd8f2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000130-vs-video_000065/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["f5171b6e-6bc8-47b6-8ffd-d8955acc9fea", "2e13e1b9-6f96-4bd2-bc7b-164be5bd329f"], "files": ["mit_franka/video_000130.mp4", "mit_franka/video_000065.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/60688e494a8dcea966ffad72.mp4"}, {"type": "video", "uri": "media/robometer_ood/c0aec9e3862f4fe7f28a8599.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000131-vs-video_000079/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["a71800dc-c84a-401c-b703-f38d6cb32389", "86ad603d-87ca-4fc9-918d-347e28f9974d"], "files": ["mit_franka/video_000079.mp4", "mit_franka/video_000131.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6fd1eb026a7026e95190cbf9.mp4"}, {"type": "video", "uri": "media/robometer_ood/52a51142a822e69700971cc0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000132-vs-video_000087/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["b0a316a8-9cad-428f-a3e6-0662ff0412dd", "196c6353-4dc2-4af0-b65c-a758e7e45345"], "files": ["mit_franka/video_000087.mp4", "mit_franka/video_000132.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2c24707ea3052389a6c0e629.mp4"}, {"type": "video", "uri": "media/robometer_ood/77ff4194d26bed19e69981d8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000133-vs-video_000090/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["d0fce1a9-ea9e-4013-bcf7-85c9f9255f2d", "74e1b687-3e35-49ac-828e-2876e978c1ac"], "files": ["mit_franka/video_000133.mp4", "mit_franka/video_000090.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2c24707ea3052389a6c0e629.mp4"}, {"type": "video", "uri": "media/robometer_ood/8fccc30dba7b991c22412112.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000133-vs-video_000104/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["d0fce1a9-ea9e-4013-bcf7-85c9f9255f2d", "42965aef-8c98-4cc9-a49c-d7f8a3a6d86e"], "files": ["mit_franka/video_000133.mp4", "mit_franka/video_000104.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9b2becb7df94f5ea088a3db6.mp4"}, {"type": "video", "uri": "media/robometer_ood/5e9767401dd7d4d67ba34dd2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000134-vs-video_000061/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["98de4790-789e-4eb0-870a-808ead842fef", "03132b75-d5b5-469c-a51d-91d2adcb2e42"], "files": ["mit_franka/video_000061.mp4", "mit_franka/video_000134.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/437a0f80d5e098ebcb7fe1d5.mp4"}, {"type": "video", "uri": "media/robometer_ood/164a0f2143f62b2e06660ed3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000136-vs-video_000069/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["ff20e79d-74a8-422d-86e6-48e3d81a5f7d", "944bc803-2cc3-459c-b340-cbdb6b54833c"], "files": ["mit_franka/video_000136.mp4", "mit_franka/video_000069.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/04bbeabac91faa36e892b1ec.mp4"}, {"type": "video", "uri": "media/robometer_ood/77cc2c0a915a86bdd1ec5e81.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000137-vs-video_000075/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["2fa9789c-8ac7-4a62-8f85-8454a1d126bd", "95359920-4e27-4363-a133-2ef7e90de9ff"], "files": ["mit_franka/video_000075.mp4", "mit_franka/video_000137.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5f5a96cbc232e6c97918cb7d.mp4"}, {"type": "video", "uri": "media/robometer_ood/1daa40e150a06f3fe9fa1498.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000138-vs-video_000093/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["41288645-ddd4-46af-9402-d67c21b98227", "2ebe5d8d-a81f-4626-95de-1941c78f3a23"], "files": ["mit_franka/video_000138.mp4", "mit_franka/video_000093.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4bc073caf2fdf83e5f723d79.mp4"}, {"type": "video", "uri": "media/robometer_ood/0c4df98d982334e81b275a46.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000142-vs-video_000095/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["14b766d5-7a47-4730-ad93-433a92162563", "01baeffb-0ef4-4921-994f-e1c52a902d09"], "files": ["mit_franka/video_000142.mp4", "mit_franka/video_000095.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3ddf31acacfc6e6da329cfa9.mp4"}, {"type": "video", "uri": "media/robometer_ood/2ec80d8c37f95668df7c231e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000143-vs-video_000099/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["b5dd5264-67f9-41c0-8aa6-07a103b19e23", "2e5d1a5a-314f-4931-acfe-8a97816ac4ce"], "files": ["mit_franka/video_000099.mp4", "mit_franka/video_000143.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f2e7820f3e1ebc1db53d9060.mp4"}, {"type": "video", "uri": "media/robometer_ood/4558f2e2282e8459e8fcde5e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000144-vs-video_000068/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["95d0a35e-ae23-4a48-bdff-00abf0713b26", "7fb935c4-ac08-44a6-b46a-d0a406e9a128"], "files": ["mit_franka/video_000144.mp4", "mit_franka/video_000068.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b172ee62fa7379adfdb739a9.mp4"}, {"type": "video", "uri": "media/robometer_ood/64f9332ff5095ea5aad18284.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000146-vs-video_000085/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["46e9bca6-7817-43fb-914b-edcc12e8010a", "a5e8ed47-92d3-4502-992e-b3c9e7fbd10c"], "files": ["mit_franka/video_000146.mp4", "mit_franka/video_000085.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6efcb2f9114ea92570301fbe.mp4"}, {"type": "video", "uri": "media/robometer_ood/19dbdfbb30f70e69675a2efd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000147-vs-video_000071/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["8ec39365-c90b-47f1-b137-fdcf28b96708", "5dba8377-0c73-48c0-89f2-85e16fee39bd"], "files": ["mit_franka/video_000071.mp4", "mit_franka/video_000147.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/19dbdfbb30f70e69675a2efd.mp4"}, {"type": "video", "uri": "media/robometer_ood/f402cebaac8d6c8e69893b47.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000147-vs-video_000105/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["5dba8377-0c73-48c0-89f2-85e16fee39bd", "52b30f87-5541-4d97-976c-f1083c06764d"], "files": ["mit_franka/video_000147.mp4", "mit_franka/video_000105.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/254bb1b49f9ee97a73aa78ba.mp4"}, {"type": "video", "uri": "media/robometer_ood/bf8ed1095bfbb822ef73a3b4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000149-vs-video_000056/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["1f348d3f-3d1c-4f2e-846e-0f085b902e9e", "4841ae7f-51c0-44d1-981c-7e22b1765ecb"], "files": ["mit_franka/video_000056.mp4", "mit_franka/video_000149.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ec9d7ce3d2589640f6970e9a.mp4"}, {"type": "video", "uri": "media/robometer_ood/c1e08aa7ceb16cbe244e48bd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000152-vs-video_000086/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["cfb656ed-cbff-4438-bf36-7193a2658da4", "20b51c66-336b-42df-a619-da72b8501547"], "files": ["mit_franka/video_000152.mp4", "mit_franka/video_000086.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ae5dd49bf572535a5de91011.mp4"}, {"type": "video", "uri": "media/robometer_ood/0d779a4c63472be0c49d3bac.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000153-vs-video_000082/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["0c6ae2b2-273d-403a-ab87-d6c5e5550f99", "566afd80-feca-456d-908f-31a64c3569c4"], "files": ["mit_franka/video_000082.mp4", "mit_franka/video_000153.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3f316f580b895ceed7921873.mp4"}, {"type": "video", "uri": "media/robometer_ood/33b18a16f1791d24c2859ddd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000155-vs-video_000094/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["f921a32c-2344-4918-8329-3a13ff0d27bf", "6a963be3-befc-454d-907a-d6634557e682"], "files": ["mit_franka/video_000155.mp4", "mit_franka/video_000094.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1dc06462371d8be6afe617f2.mp4"}, {"type": "video", "uri": "media/robometer_ood/fc5bafe3320b30dd6fd89aff.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000162-vs-video_000078/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["22bde85d-a824-42c1-8c11-3ce96dc30b29", "5a7d93e2-296f-4dd0-a804-4d11a5c2f0bf"], "files": ["mit_franka/video_000162.mp4", "mit_franka/video_000078.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c4b93f42530c139811a3f8d0.mp4"}, {"type": "video", "uri": "media/robometer_ood/0c6ceecef563b31cf2939f91.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000163-vs-video_000097/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["6d6781e8-d812-4968-8e1f-8ae633df7341", "2f3fb3c1-4d80-4737-8da3-6113c4566ab6"], "files": ["mit_franka/video_000163.mp4", "mit_franka/video_000097.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/452813b5e62035853fc63b30.mp4"}, {"type": "video", "uri": "media/robometer_ood/79ea9d39608f8f87e556ad57.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000167-vs-video_000088/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["2da3ac9d-c65a-4b97-af12-4836a9a06c0d", "800f0991-83b9-477a-b883-2b7abf401534"], "files": ["mit_franka/video_000167.mp4", "mit_franka/video_000088.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/44988f4bdfd0ad1b87ff18e5.mp4"}, {"type": "video", "uri": "media/robometer_ood/d405875905853e48957e2fd2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000169-vs-video_000077/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["fd03f7c8-1fbb-4095-9fd0-58ddf8ab10f7", "aced9fe0-caee-4790-b4b7-e4ef20febf1b"], "files": ["mit_franka/video_000077.mp4", "mit_franka/video_000169.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7b67c18daba4f902ef0d217b.mp4"}, {"type": "video", "uri": "media/robometer_ood/88ddc8c392c19c2393bd3a64.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000170-vs-video_000059/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["02d6f4da-fee8-4362-807f-e90dcedc3c20", "23c242da-0437-4224-aeb5-9d2d021d40a6"], "files": ["mit_franka/video_000059.mp4", "mit_franka/video_000170.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/88ddc8c392c19c2393bd3a64.mp4"}, {"type": "video", "uri": "media/robometer_ood/fa50f8ca47ce609f0ae567f9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000170-vs-video_000096/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["23c242da-0437-4224-aeb5-9d2d021d40a6", "3677a3d6-b3b4-485e-a126-f6bb39b2ddbc"], "files": ["mit_franka/video_000170.mp4", "mit_franka/video_000096.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7c422f3ac169369626fd99ba.mp4"}, {"type": "video", "uri": "media/robometer_ood/f660e59eae13450d9af876f8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000172-vs-video_000081/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["e8ae15de-eb7d-40cc-a504-b66f8ff602b4", "0c1d3d91-a9f0-423e-af58-34d61b90f7cd"], "files": ["mit_franka/video_000081.mp4", "mit_franka/video_000172.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/32e0d9eceb57d63eb6ea49d4.mp4"}, {"type": "video", "uri": "media/robometer_ood/32f9ebfc6076e809ba0809f3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000174-vs-video_000101/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["fa85c086-d384-4e53-a2d6-a7286b86e5c7", "7e4c9f98-39c1-4984-b86b-4f60d9eec5a5"], "files": ["mit_franka/video_000174.mp4", "mit_franka/video_000101.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4ad7d98e9dffb3248d9d7230.mp4"}, {"type": "video", "uri": "media/robometer_ood/39e69f102052110a55dbd2dd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000176-vs-video_000083/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["3b289cc8-0e0d-4431-a9e8-685430eac45b", "061f7ba2-297c-4ad3-aac5-a49e9aae9a16"], "files": ["mit_franka/video_000176.mp4", "mit_franka/video_000083.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/85efd00b00b26190ea099729.mp4"}, {"type": "video", "uri": "media/robometer_ood/9e32266c14919f75e7d40252.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000180-vs-video_000098/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["8711a72c-b077-487d-82f5-eb271a00e839", "d929ea18-4f95-4015-8804-919dd091c626"], "files": ["mit_franka/video_000098.mp4", "mit_franka/video_000180.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6cbfc6d0b4762799d1dc2767.mp4"}, {"type": "video", "uri": "media/robometer_ood/f667ff391a6dc25fd532fac5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000182-vs-video_000091/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["e495eabb-9540-4cf3-a8ed-f850a7c4d979", "eb51e61f-1866-4fb9-814a-12b2f5a44298"], "files": ["mit_franka/video_000182.mp4", "mit_franka/video_000091.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ce467e9743283e54135036b5.mp4"}, {"type": "video", "uri": "media/robometer_ood/c8ef6a59e0b8f559f8a14ec6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000186-vs-video_000058/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["fd2789d8-d799-4fdf-9d09-a9c4ff3355ed", "eea29928-bb0a-4e0a-9e92-7765a60f569b"], "files": ["mit_franka/video_000058.mp4", "mit_franka/video_000186.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bc8c611a5c6a87e493c74aa0.mp4"}, {"type": "video", "uri": "media/robometer_ood/7b42da0ec3bce19dd7f03209.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000187-vs-video_000064/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["8122a98a-2464-4d79-a555-b0857ab208ad", "8d857e6b-d6a6-4799-ac84-57d854b1c353"], "files": ["mit_franka/video_000064.mp4", "mit_franka/video_000187.mp4"], "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 163, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b6399de0395a4f76bc3cd2ed.mp4"}, {"type": "video", "uri": "media/robometer_ood/1a8cd1259bea5596a0c3b458.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000188-vs-video_000203/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["4ddc02f8-c62c-4e6d-938c-4677fbfb7afb", "29dc44e8-3ae1-4849-aa0f-8efb7e2c6e04"], "files": ["mit_franka/video_000188.mp4", "mit_franka/video_000203.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ad18f02a835323b431fd4b28.mp4"}, {"type": "video", "uri": "media/robometer_ood/e71bf028974f5f49a3323ef0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000189-vs-video_000199/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["a8c9b1ae-f965-4445-8792-3a7b11e0f504", "7f865ab8-c7c9-47f9-b7a9-a1dcf3bf5f41"], "files": ["mit_franka/video_000189.mp4", "mit_franka/video_000199.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b9583801513df4fafb9e0d8c.mp4"}, {"type": "video", "uri": "media/robometer_ood/8cc26d98b31575dfbb0a4cfd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000201-vs-video_000194/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["5f4ce0ee-ac41-4d54-bb09-903f1153c78c", "dfe31e07-ba00-418e-bc78-8fd67141fce2"], "files": ["mit_franka/video_000201.mp4", "mit_franka/video_000194.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e71bf028974f5f49a3323ef0.mp4"}, {"type": "video", "uri": "media/robometer_ood/7a600269b3d684a78718d131.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000202-vs-video_000199/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["7f865ab8-c7c9-47f9-b7a9-a1dcf3bf5f41", "ab343a28-eb35-4e31-99b9-ba5fa8225d79"], "files": ["mit_franka/video_000199.mp4", "mit_franka/video_000202.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a651aabbd84818fb477e6b98.mp4"}, {"type": "video", "uri": "media/robometer_ood/0855f40ebcc8a87d8eff1957.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000204-vs-video_000193/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["2f98b96c-ea1d-419f-bda1-e5af330d80f6", "94a8857c-bbaf-4d02-9582-368991ecc094"], "files": ["mit_franka/video_000193.mp4", "mit_franka/video_000204.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4c3b67484b6f5f9dea6ce75d.mp4"}, {"type": "video", "uri": "media/robometer_ood/b6399de0395a4f76bc3cd2ed.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000205-vs-video_000188/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["67ea3c81-d7e1-4ddb-840e-4f16e09a8da2", "4ddc02f8-c62c-4e6d-938c-4677fbfb7afb"], "files": ["mit_franka/video_000205.mp4", "mit_franka/video_000188.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4c3b67484b6f5f9dea6ce75d.mp4"}, {"type": "video", "uri": "media/robometer_ood/e1658bb8e1395290120d16de.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000205-vs-video_000195/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["67ea3c81-d7e1-4ddb-840e-4f16e09a8da2", "70d81e02-e13b-40ca-ba35-0145cc226746"], "files": ["mit_franka/video_000205.mp4", "mit_franka/video_000195.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f32c6696483e20a0bd5e6b0e.mp4"}, {"type": "video", "uri": "media/robometer_ood/87fda6de370b6301a419cbe5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000206-vs-video_000197/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["15f075e4-d639-4a40-949e-199a0e5dafcd", "b9a63762-7e22-4194-85a2-93d5666d876c"], "files": ["mit_franka/video_000206.mp4", "mit_franka/video_000197.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c8bb536df9c4f5be6fcb5cde.mp4"}, {"type": "video", "uri": "media/robometer_ood/ad18f02a835323b431fd4b28.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000207-vs-video_000189/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["0e067172-7a09-42b1-a71a-2ff359818496", "a8c9b1ae-f965-4445-8792-3a7b11e0f504"], "files": ["mit_franka/video_000207.mp4", "mit_franka/video_000189.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/de9d657d67e76259ab2c1a27.mp4"}, {"type": "video", "uri": "media/robometer_ood/c8bb536df9c4f5be6fcb5cde.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000207-vs-video_000200/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["d2130229-c90e-4f99-89a8-708ee3d7d40c", "0e067172-7a09-42b1-a71a-2ff359818496"], "files": ["mit_franka/video_000200.mp4", "mit_franka/video_000207.mp4"], "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 166, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b8df5dcbdc6e312dfbee4c2e.mp4"}, {"type": "video", "uri": "media/robometer_ood/3c3c115fb1f73ef9cfd87f35.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000208-vs-video_000219/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["5a977a2c-6b2d-4c3d-9266-44c66d77f666", "358c22bf-26d1-4bee-9a30-5021f012c297"], "files": ["mit_franka/video_000219.mp4", "mit_franka/video_000208.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/12b0dfeb26ee3fd421ec94b2.mp4"}, {"type": "video", "uri": "media/robometer_ood/77c08688fe06c45970b22d65.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000209-vs-video_000224/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["37cba8cf-f5d7-4be2-80bb-749cd6b40afb", "59e8ecf8-9f31-404f-a3d8-04d987a24a08"], "files": ["mit_franka/video_000224.mp4", "mit_franka/video_000209.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d87546b33f7825e8ee7e92e5.mp4"}, {"type": "video", "uri": "media/robometer_ood/ac6fe3106ba76199791a7df8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000210-vs-video_000227/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["49c291d4-ca32-43b0-ab1d-18971e1cc5ba", "ff14061e-f0f0-4ddc-936c-bad67b5db46d"], "files": ["mit_franka/video_000227.mp4", "mit_franka/video_000210.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/dd8366de3f7d8291d7a0a968.mp4"}, {"type": "video", "uri": "media/robometer_ood/c6290fc3f924e8ff889b2d7f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000211-vs-video_000231/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["ef188e75-850d-48e2-9d08-85aa65366970", "ccb4bc2e-7531-4c0a-b007-62967a77628d"], "files": ["mit_franka/video_000211.mp4", "mit_franka/video_000231.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/09b487354d94407fe56de0dd.mp4"}, {"type": "video", "uri": "media/robometer_ood/d05341995ab357dc20eee8d3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000212-vs-video_000222/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["becf8d17-d42d-41c6-8929-2a05f270ccf9", "5920cd7a-c2b3-47ab-bc88-79b18e1024fb"], "files": ["mit_franka/video_000212.mp4", "mit_franka/video_000222.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9d2fbbb80fd036bc3984aa6e.mp4"}, {"type": "video", "uri": "media/robometer_ood/213d04898141187fc4cc6686.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000213-vs-video_000221/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["c3938b4a-0d7a-4df2-82ab-397b201449dd", "4ffef4c9-d1d1-457e-bd52-6d259faa6ae4"], "files": ["mit_franka/video_000221.mp4", "mit_franka/video_000213.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/23d61a3044e7b34a3457d312.mp4"}, {"type": "video", "uri": "media/robometer_ood/c732e3e0ac12c6533694ae02.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000216-vs-video_000233/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["5bdaf18f-3af6-43a2-b3ed-76a519dfef0f", "55b8baa3-0054-4abe-acba-b3c66577d4fd"], "files": ["mit_franka/video_000216.mp4", "mit_franka/video_000233.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1a83d9851e7c84df7a778507.mp4"}, {"type": "video", "uri": "media/robometer_ood/3a9a286dcca6c40c264e6c91.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000217-vs-video_000226/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["b120cfab-87c7-4733-88d7-563bedc816e1", "60dd34a2-a44b-4f7b-b97a-9189ec164365"], "files": ["mit_franka/video_000217.mp4", "mit_franka/video_000226.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1a83d9851e7c84df7a778507.mp4"}, {"type": "video", "uri": "media/robometer_ood/9f4ce3e469a66c7be3a42703.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000234-vs-video_000217/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["b120cfab-87c7-4733-88d7-563bedc816e1", "2bc5e41a-f2f9-469e-9c61-edea17686a67"], "files": ["mit_franka/video_000217.mp4", "mit_franka/video_000234.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9f4ce3e469a66c7be3a42703.mp4"}, {"type": "video", "uri": "media/robometer_ood/3aa0f97b384ecb637d4befe5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000234-vs-video_000220/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["2bc5e41a-f2f9-469e-9c61-edea17686a67", "a75c2cac-349d-4d15-93ce-ad2f1f7877bc"], "files": ["mit_franka/video_000234.mp4", "mit_franka/video_000220.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/38288578b72cc496250b0d94.mp4"}, {"type": "video", "uri": "media/robometer_ood/77c08688fe06c45970b22d65.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000235-vs-video_000209/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["d4bd0b33-6be2-4276-be13-28861e9ef0ee", "59e8ecf8-9f31-404f-a3d8-04d987a24a08"], "files": ["mit_franka/video_000235.mp4", "mit_franka/video_000209.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/38288578b72cc496250b0d94.mp4"}, {"type": "video", "uri": "media/robometer_ood/5136eab7c5f0c69ef0ce2a05.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000235-vs-video_000228/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["d4bd0b33-6be2-4276-be13-28861e9ef0ee", "4d30a0de-bd76-43fb-ab53-42548d043162"], "files": ["mit_franka/video_000235.mp4", "mit_franka/video_000228.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8d0c7fc70a510c8827bc9464.mp4"}, {"type": "video", "uri": "media/robometer_ood/09b487354d94407fe56de0dd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000236-vs-video_000212/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["10ffeb3c-19b2-400a-9112-570fcaa5902a", "becf8d17-d42d-41c6-8929-2a05f270ccf9"], "files": ["mit_franka/video_000236.mp4", "mit_franka/video_000212.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c732e3e0ac12c6533694ae02.mp4"}, {"type": "video", "uri": "media/robometer_ood/8d0c7fc70a510c8827bc9464.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000236-vs-video_000233/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["55b8baa3-0054-4abe-acba-b3c66577d4fd", "10ffeb3c-19b2-400a-9112-570fcaa5902a"], "files": ["mit_franka/video_000233.mp4", "mit_franka/video_000236.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/20ab7299ba757374dbe463e7.mp4"}, {"type": "video", "uri": "media/robometer_ood/fbed8695daebe66024ce7fce.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000237-vs-video_000255/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["362444b2-3b7e-4e0a-b025-23e6071c0944", "0df2ebeb-cf62-429b-a703-16eaa0b840da"], "files": ["mit_franka/video_000255.mp4", "mit_franka/video_000237.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8dfbe5121d43ef5a154f323d.mp4"}, {"type": "video", "uri": "media/robometer_ood/3c3c115fb1f73ef9cfd87f35.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000238-vs-video_000208/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["f1a636c7-5760-4b3e-bdf4-9bd003f0172d", "358c22bf-26d1-4bee-9a30-5021f012c297"], "files": ["mit_franka/video_000238.mp4", "mit_franka/video_000208.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d87546b33f7825e8ee7e92e5.mp4"}, {"type": "video", "uri": "media/robometer_ood/8dfbe5121d43ef5a154f323d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000238-vs-video_000227/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["49c291d4-ca32-43b0-ab1d-18971e1cc5ba", "f1a636c7-5760-4b3e-bdf4-9bd003f0172d"], "files": ["mit_franka/video_000227.mp4", "mit_franka/video_000238.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c4b2424f70d2c53fb6fafa54.mp4"}, {"type": "video", "uri": "media/robometer_ood/30420f2de5f98c011bacf360.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000239-vs-video_000249/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["588e3fa8-039d-444b-bcf0-bd39cecd436c", "1f545ea7-a5c1-4c70-9969-1466c85bdd3e"], "files": ["mit_franka/video_000249.mp4", "mit_franka/video_000239.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0aa734a9dd55e033f022aebc.mp4"}, {"type": "video", "uri": "media/robometer_ood/dd8366de3f7d8291d7a0a968.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000240-vs-video_000211/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["132839cd-841c-480e-940c-9c22d34dd355", "ef188e75-850d-48e2-9d08-85aa65366970"], "files": ["mit_franka/video_000240.mp4", "mit_franka/video_000211.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0aa734a9dd55e033f022aebc.mp4"}, {"type": "video", "uri": "media/robometer_ood/d05341995ab357dc20eee8d3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000240-vs-video_000222/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["132839cd-841c-480e-940c-9c22d34dd355", "5920cd7a-c2b3-47ab-bc88-79b18e1024fb"], "files": ["mit_franka/video_000240.mp4", "mit_franka/video_000222.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/26580546b3e49ad9736ab422.mp4"}, {"type": "video", "uri": "media/robometer_ood/f0708f74dc95d3c9ea383e8f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000241-vs-video_000250/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["91dcc5c5-1b64-4222-ae15-45ddbff63d64", "654fad87-0297-41da-86cd-513afa7ae08f"], "files": ["mit_franka/video_000250.mp4", "mit_franka/video_000241.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/72382037050996d7c82495a8.mp4"}, {"type": "video", "uri": "media/robometer_ood/5e36d20c3bb99e9ba5144453.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000242-vs-video_000261/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["b9d0e32f-df63-49b7-80cf-96feb65677ce", "c5867dae-2609-4a07-b3cf-603042618ba7"], "files": ["mit_franka/video_000261.mp4", "mit_franka/video_000242.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8b7c019829cb4c59eb66917e.mp4"}, {"type": "video", "uri": "media/robometer_ood/8dd42d03ed62ecb774ee890d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000243-vs-video_000251/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["4c9422ed-3f72-4eb6-a7eb-feea7c18e696", "0b06aa82-902d-48aa-a4c3-82b12071c597"], "files": ["mit_franka/video_000251.mp4", "mit_franka/video_000243.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/23d61a3044e7b34a3457d312.mp4"}, {"type": "video", "uri": "media/robometer_ood/24099e99fe0117887431dce3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000244-vs-video_000216/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["5bdaf18f-3af6-43a2-b3ed-76a519dfef0f", "b00d51d5-9ad6-45cf-aca8-03df33c632e7"], "files": ["mit_franka/video_000216.mp4", "mit_franka/video_000244.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/24099e99fe0117887431dce3.mp4"}, {"type": "video", "uri": "media/robometer_ood/c6290fc3f924e8ff889b2d7f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000244-vs-video_000231/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["b00d51d5-9ad6-45cf-aca8-03df33c632e7", "ccb4bc2e-7531-4c0a-b007-62967a77628d"], "files": ["mit_franka/video_000244.mp4", "mit_franka/video_000231.mp4"], "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 159, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/77ee83ea30b6c17a0396d9f7.mp4"}, {"type": "video", "uri": "media/robometer_ood/a6a5ccab792ec11c7536a8b2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000245-vs-video_000247/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["80053f52-e509-48e2-b6fd-b426fb349055", "92704c22-37e3-47d7-8d98-f9a2660e59e1"], "files": ["mit_franka/video_000245.mp4", "mit_franka/video_000247.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4d253d67b7436b1c1ea4dd82.mp4"}, {"type": "video", "uri": "media/robometer_ood/5e36d20c3bb99e9ba5144453.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000254-vs-video_000242/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["37c316ad-2fbf-47de-a713-bd63d7148a00", "c5867dae-2609-4a07-b3cf-603042618ba7"], "files": ["mit_franka/video_000254.mp4", "mit_franka/video_000242.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/26580546b3e49ad9736ab422.mp4"}, {"type": "video", "uri": "media/robometer_ood/4d253d67b7436b1c1ea4dd82.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000254-vs-video_000250/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["91dcc5c5-1b64-4222-ae15-45ddbff63d64", "37c316ad-2fbf-47de-a713-bd63d7148a00"], "files": ["mit_franka/video_000250.mp4", "mit_franka/video_000254.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c659f9ba2b71317c04a9dbea.mp4"}, {"type": "video", "uri": "media/robometer_ood/f0708f74dc95d3c9ea383e8f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000256-vs-video_000241/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["fd33b95b-e031-4057-9356-453c63fa4f8d", "654fad87-0297-41da-86cd-513afa7ae08f"], "files": ["mit_franka/video_000256.mp4", "mit_franka/video_000241.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/72382037050996d7c82495a8.mp4"}, {"type": "video", "uri": "media/robometer_ood/c659f9ba2b71317c04a9dbea.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000256-vs-video_000261/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["b9d0e32f-df63-49b7-80cf-96feb65677ce", "fd33b95b-e031-4057-9356-453c63fa4f8d"], "files": ["mit_franka/video_000261.mp4", "mit_franka/video_000256.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e5b78e19862640e79c9406be.mp4"}, {"type": "video", "uri": "media/robometer_ood/8dd42d03ed62ecb774ee890d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000257-vs-video_000243/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["cdbebfe5-2858-49e1-b807-e88935c2f003", "0b06aa82-902d-48aa-a4c3-82b12071c597"], "files": ["mit_franka/video_000257.mp4", "mit_franka/video_000243.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/46545eec355dab9c311e51fd.mp4"}, {"type": "video", "uri": "media/robometer_ood/e5b78e19862640e79c9406be.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000257-vs-video_000253/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["774e444f-681c-4602-9c1a-c87c59cb9e3f", "cdbebfe5-2858-49e1-b807-e88935c2f003"], "files": ["mit_franka/video_000253.mp4", "mit_franka/video_000257.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6b0df54159fbd088f242596b.mp4"}, {"type": "video", "uri": "media/robometer_ood/fbed8695daebe66024ce7fce.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000258-vs-video_000237/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["23667b25-131a-44c6-aa31-9413c4a18a7a", "0df2ebeb-cf62-429b-a703-16eaa0b840da"], "files": ["mit_franka/video_000258.mp4", "mit_franka/video_000237.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d53668fc7bd1cad5cae26103.mp4"}, {"type": "video", "uri": "media/robometer_ood/6b0df54159fbd088f242596b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000258-vs-video_000246/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["5293dca8-22bd-49e0-bf4a-014b4a87fc42", "23667b25-131a-44c6-aa31-9413c4a18a7a"], "files": ["mit_franka/video_000246.mp4", "mit_franka/video_000258.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/30420f2de5f98c011bacf360.mp4"}, {"type": "video", "uri": "media/robometer_ood/ae06bdaf9ffc19b43d7d2a9e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000259-vs-video_000239/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["1f545ea7-a5c1-4c70-9969-1466c85bdd3e", "f941604f-ad46-43ad-914a-2c53eef0bd8c"], "files": ["mit_franka/video_000239.mp4", "mit_franka/video_000259.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/97807926afc896668ad66cf2.mp4"}, {"type": "video", "uri": "media/robometer_ood/ae06bdaf9ffc19b43d7d2a9e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000259-vs-video_000252/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["b428c0c7-80ed-4dd8-8c42-8000bea2fd78", "f941604f-ad46-43ad-914a-2c53eef0bd8c"], "files": ["mit_franka/video_000252.mp4", "mit_franka/video_000259.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/77ee83ea30b6c17a0396d9f7.mp4"}, {"type": "video", "uri": "media/robometer_ood/90d78fe0a76563fd90fa0a0b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000260-vs-video_000245/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["80053f52-e509-48e2-b6fd-b426fb349055", "23c5eeec-16a3-4cbd-9614-ea3ae9a47771"], "files": ["mit_franka/video_000245.mp4", "mit_franka/video_000260.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pull a tissue from the tissue box", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/90d78fe0a76563fd90fa0a0b.mp4"}, {"type": "video", "uri": "media/robometer_ood/20ab7299ba757374dbe463e7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000260-vs-video_000255/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["23c5eeec-16a3-4cbd-9614-ea3ae9a47771", "362444b2-3b7e-4e0a-b025-23e6071c0944"], "files": ["mit_franka/video_000260.mp4", "mit_franka/video_000255.mp4"], "task_norm": "Pull a tissue from the tissue box", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/905741e54adfb3b593f325b5.mp4"}, {"type": "video", "uri": "media/robometer_ood/ddb4ac46a1ec6a448ad21614.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000263-vs-video_000274/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["de9478ee-fc31-47f5-a699-5f024e57dbc3", "0151e02c-f5c5-43a6-be60-3bc377b6500f"], "files": ["mit_franka/video_000263.mp4", "mit_franka/video_000274.mp4"], "task_norm": "Pick up the spoon and place it inside the cup", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/86f901e15f60d7ac5fb5e3a9.mp4"}, {"type": "video", "uri": "media/robometer_ood/4726439bc88057002ca58f08.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000264-vs-video_000272/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["d4b21264-5347-43db-99a0-1aa67d59dae2", "769dfeea-2152-4c6e-b947-d4209c4fc407"], "files": ["mit_franka/video_000272.mp4", "mit_franka/video_000264.mp4"], "task_norm": "Pick up the spoon and place it inside the cup", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/69923323bd9926e2825bf4c6.mp4"}, {"type": "video", "uri": "media/robometer_ood/71a5b760baa26925d6722806.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000265-vs-video_000276/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["287477a7-8b5d-437e-b589-fe776174f251", "95bef17e-ae03-4040-ac10-c15c01e2cce3"], "files": ["mit_franka/video_000265.mp4", "mit_franka/video_000276.mp4"], "task_norm": "Pick up the spoon and place it inside the cup", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/636cda7938d7029aef72de4c.mp4"}, {"type": "video", "uri": "media/robometer_ood/84464a8110c44b6e63041e00.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000270-vs-video_000273/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["4dd57b74-112b-4241-b941-e1530d3d93b9", "382c8174-8634-4b8f-9c70-d47a7b76e40f"], "files": ["mit_franka/video_000273.mp4", "mit_franka/video_000270.mp4"], "task_norm": "Pick up the spoon and place it inside the cup", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/69923323bd9926e2825bf4c6.mp4"}, {"type": "video", "uri": "media/robometer_ood/9dfb9c1db36187bf5ca0a91c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000275-vs-video_000265/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["287477a7-8b5d-437e-b589-fe776174f251", "12c4b30a-4d5c-47cf-99da-aa4cd73433f7"], "files": ["mit_franka/video_000265.mp4", "mit_franka/video_000275.mp4"], "task_norm": "Pick up the spoon and place it inside the cup", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ddb4ac46a1ec6a448ad21614.mp4"}, {"type": "video", "uri": "media/robometer_ood/9dfb9c1db36187bf5ca0a91c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000275-vs-video_000274/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["0151e02c-f5c5-43a6-be60-3bc377b6500f", "12c4b30a-4d5c-47cf-99da-aa4cd73433f7"], "files": ["mit_franka/video_000274.mp4", "mit_franka/video_000275.mp4"], "task_norm": "Pick up the spoon and place it inside the cup", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a8144787ad3e3abfb17fb317.mp4"}, {"type": "video", "uri": "media/robometer_ood/ab3915a2aed9862634ae744c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000277-vs-video_000288/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["2659ecbf-dee3-4ef8-b32c-aa4748ddb236", "d2d0d2e1-536c-4ea7-888f-fa5745aae2d1"], "files": ["mit_franka/video_000277.mp4", "mit_franka/video_000288.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e5c5bae4f1a7cc90407c0679.mp4"}, {"type": "video", "uri": "media/robometer_ood/792cec0ad8f76dbc231baba7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000278-vs-video_000269/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["bd5008fe-b0ff-4d7b-b994-b7435b756829", "61074f93-9e7c-4d2d-9c9e-d5f2f9e2f386"], "files": ["mit_franka/video_000278.mp4", "mit_franka/video_000269.mp4"], "task_norm": "Pick up the spoon and place it inside the cup", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e5c5bae4f1a7cc90407c0679.mp4"}, {"type": "video", "uri": "media/robometer_ood/636cda7938d7029aef72de4c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000278-vs-video_000273/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["bd5008fe-b0ff-4d7b-b994-b7435b756829", "4dd57b74-112b-4241-b941-e1530d3d93b9"], "files": ["mit_franka/video_000278.mp4", "mit_franka/video_000273.mp4"], "task_norm": "Pick up the spoon and place it inside the cup", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2a974a83cd564e6b6fee7dc1.mp4"}, {"type": "video", "uri": "media/robometer_ood/53f9538a1c3a9f8e0beaebb9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000279-vs-video_000291/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["9d084f85-1caf-47eb-86e1-6e04360bbcf5", "65fba2df-dba3-42c8-b1a8-53d05e9f3f55"], "files": ["mit_franka/video_000291.mp4", "mit_franka/video_000279.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/57197b4367750e1e5be61da2.mp4"}, {"type": "video", "uri": "media/robometer_ood/4e3a7da83c4601f82619a043.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000283-vs-video_000286/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["533a2da5-3347-4111-a826-98fac3940ecb", "9aa6ff0d-a0c0-4952-9faf-14df41ef387a"], "files": ["mit_franka/video_000283.mp4", "mit_franka/video_000286.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/64e83bf364e9b8e000c74091.mp4"}, {"type": "video", "uri": "media/robometer_ood/9cf84fe2476d9ca4943f9ff6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000285-vs-video_000293/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["13fe84df-e15f-46b5-b912-7469492471cd", "2680f750-fa8c-48ab-9509-cc6487e0f3d0"], "files": ["mit_franka/video_000293.mp4", "mit_franka/video_000285.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1ca4e8fa17bd44b9e27b74be.mp4"}, {"type": "video", "uri": "media/robometer_ood/53f9538a1c3a9f8e0beaebb9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000296-vs-video_000279/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["87c70b30-5a99-4411-9add-7790f2d5a297", "65fba2df-dba3-42c8-b1a8-53d05e9f3f55"], "files": ["mit_franka/video_000296.mp4", "mit_franka/video_000279.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1ca4e8fa17bd44b9e27b74be.mp4"}, {"type": "video", "uri": "media/robometer_ood/5dd97b579eef30f50b992845.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000296-vs-video_000292/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["87c70b30-5a99-4411-9add-7790f2d5a297", "36f6efbe-618d-4f8a-9248-dfd418f622b2"], "files": ["mit_franka/video_000296.mp4", "mit_franka/video_000292.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4beffc864b4f0a0d205df597.mp4"}, {"type": "video", "uri": "media/robometer_ood/9f01e05b635028e6ef70ca5f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000297-vs-video_000281/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["07ba1444-4ea6-45c7-9b2c-3b74a1fdb49f", "5b261f71-de3a-4626-bb12-6aa2f69af2f0"], "files": ["mit_franka/video_000297.mp4", "mit_franka/video_000281.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ea1437d0686506166f2a01e9.mp4"}, {"type": "video", "uri": "media/robometer_ood/612cffea073a7cccf6f2e3dc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000298-vs-video_000282/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["8d1c9687-085e-40e3-b7ab-e693d917f020", "aeeb8a39-2eff-4ce2-95da-b3dcf51a7346"], "files": ["mit_franka/video_000298.mp4", "mit_franka/video_000282.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9cf84fe2476d9ca4943f9ff6.mp4"}, {"type": "video", "uri": "media/robometer_ood/25ae0ebdbab12845d3a1b13e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000299-vs-video_000285/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["2680f750-fa8c-48ab-9509-cc6487e0f3d0", "03249d25-4221-4b43-bc21-c72bb3846592"], "files": ["mit_franka/video_000285.mp4", "mit_franka/video_000299.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/25ae0ebdbab12845d3a1b13e.mp4"}, {"type": "video", "uri": "media/robometer_ood/ab3915a2aed9862634ae744c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000299-vs-video_000288/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["03249d25-4221-4b43-bc21-c72bb3846592", "d2d0d2e1-536c-4ea7-888f-fa5745aae2d1"], "files": ["mit_franka/video_000299.mp4", "mit_franka/video_000288.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e72fac9a67a94cd35ddcf5dc.mp4"}, {"type": "video", "uri": "media/robometer_ood/80f6e3eda8135eca98435b88.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000300-vs-video_000294/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["69bb936b-5edd-4ce4-b0ee-0b8014c39137", "1d9930d7-eb19-4d96-96ba-f3fdcdb2fe72"], "files": ["mit_franka/video_000294.mp4", "mit_franka/video_000300.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a8144787ad3e3abfb17fb317.mp4"}, {"type": "video", "uri": "media/robometer_ood/3683bfaa4ded976bc17711d5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000301-vs-video_000277/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["2659ecbf-dee3-4ef8-b32c-aa4748ddb236", "6d5c218b-66cd-4395-94c5-4024cde9f93c"], "files": ["mit_franka/video_000277.mp4", "mit_franka/video_000301.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3683bfaa4ded976bc17711d5.mp4"}, {"type": "video", "uri": "media/robometer_ood/3d93eed3da94339b79c7704c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000301-vs-video_000290/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["6d5c218b-66cd-4395-94c5-4024cde9f93c", "ea0b24b3-a13d-4b1d-9fcb-4641918adde3"], "files": ["mit_franka/video_000301.mp4", "mit_franka/video_000290.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f0e01e147c1a50a6e93ced53.mp4"}, {"type": "video", "uri": "media/robometer_ood/2a974a83cd564e6b6fee7dc1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000302-vs-video_000291/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["36585075-3c6d-4f06-b112-a90f84556cb4", "9d084f85-1caf-47eb-86e1-6e04360bbcf5"], "files": ["mit_franka/video_000302.mp4", "mit_franka/video_000291.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/57197b4367750e1e5be61da2.mp4"}, {"type": "video", "uri": "media/robometer_ood/f7d04c699265a3d26f045692.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000303-vs-video_000283/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["533a2da5-3347-4111-a826-98fac3940ecb", "48279aac-ff7f-462d-9a1d-9ab718f8c2dd"], "files": ["mit_franka/video_000283.mp4", "mit_franka/video_000303.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f7d04c699265a3d26f045692.mp4"}, {"type": "video", "uri": "media/robometer_ood/780df5e82b93095de5542db8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000303-vs-video_000284/robometer_traj_pref", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["48279aac-ff7f-462d-9a1d-9ab718f8c2dd", "26242dd7-a9a6-4290-bb00-30f206f97876"], "files": ["mit_franka/video_000303.mp4", "mit_franka/video_000284.mp4"], "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": [{"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 160, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cc4203dec13fcbf9a2ea0a01.mp4"}, {"type": "video", "uri": "media/robometer_ood/05c4bea2e5254a592770680c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000002-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["d6b86600-6750-4e21-9892-b67442c97841", "17a39657-d229-4003-8147-0c04fadc9bb1"], "files": ["usc_franka/video_000000.mp4", "usc_franka/video_000002.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ffa5e3e9d24379327829bf4b.mp4"}, {"type": "video", "uri": "media/robometer_ood/0072dce5a2e0296aa21fd341.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000003-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["ba36cce3-cc6f-44bc-8746-0a777b00dd5b", "fd70a190-7043-4a8e-ac24-c3ea943856e5"], "files": ["usc_franka/video_000003.mp4", "usc_franka/video_000001.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cc4203dec13fcbf9a2ea0a01.mp4"}, {"type": "video", "uri": "media/robometer_ood/f01f73e81747b54c3dfec63e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000004-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["d6b86600-6750-4e21-9892-b67442c97841", "da0ecbfe-d086-4a01-857f-d0cdb9e30b72"], "files": ["usc_franka/video_000000.mp4", "usc_franka/video_000004.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ffa5e3e9d24379327829bf4b.mp4"}, {"type": "video", "uri": "media/robometer_ood/f01f73e81747b54c3dfec63e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000004-vs-video_000003/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["ba36cce3-cc6f-44bc-8746-0a777b00dd5b", "da0ecbfe-d086-4a01-857f-d0cdb9e30b72"], "files": ["usc_franka/video_000003.mp4", "usc_franka/video_000004.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0072dce5a2e0296aa21fd341.mp4"}, {"type": "video", "uri": "media/robometer_ood/1fb3d361c7d4d9fdd6c6cc90.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000005-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["fd70a190-7043-4a8e-ac24-c3ea943856e5", "d8a0dba2-e4b3-4db3-bce7-60538500224d"], "files": ["usc_franka/video_000001.mp4", "usc_franka/video_000005.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/05c4bea2e5254a592770680c.mp4"}, {"type": "video", "uri": "media/robometer_ood/1fb3d361c7d4d9fdd6c6cc90.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000005-vs-video_000002/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["17a39657-d229-4003-8147-0c04fadc9bb1", "d8a0dba2-e4b3-4db3-bce7-60538500224d"], "files": ["usc_franka/video_000002.mp4", "usc_franka/video_000005.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Put the yellow plate in the sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/507c44d2bb43eaad67947b5d.mp4"}, {"type": "video", "uri": "media/robometer_ood/1d7e7efe744ea7801218e8e7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000008-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["c2a93b8e-5c54-4472-8ed9-d7abee1338fd", "6f1c7142-25fe-4fd3-b6ff-7b1488a2aec1"], "files": ["usc_franka/video_000007.mp4", "usc_franka/video_000008.mp4"], "task_norm": "Put the yellow plate in the sink", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the yellow plate in the sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8555d91672ea243df982509b.mp4"}, {"type": "video", "uri": "media/robometer_ood/169dd0a1a099a820b606574e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000009-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["9433cc45-8f7b-4b7c-9567-4e51a372cc55", "94af8ff4-0f77-48fd-9a83-cc35fd060e66"], "files": ["usc_franka/video_000009.mp4", "usc_franka/video_000006.mp4"], "task_norm": "Put the yellow plate in the sink", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the yellow plate in the sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/507c44d2bb43eaad67947b5d.mp4"}, {"type": "video", "uri": "media/robometer_ood/0b704cd3b1f13c7e609a97d4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000010-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["c2a93b8e-5c54-4472-8ed9-d7abee1338fd", "4a7082b7-018e-49cd-8484-ab15695a71e9"], "files": ["usc_franka/video_000007.mp4", "usc_franka/video_000010.mp4"], "task_norm": "Put the yellow plate in the sink", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the yellow plate in the sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8555d91672ea243df982509b.mp4"}, {"type": "video", "uri": "media/robometer_ood/0b704cd3b1f13c7e609a97d4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000010-vs-video_000009/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["9433cc45-8f7b-4b7c-9567-4e51a372cc55", "4a7082b7-018e-49cd-8484-ab15695a71e9"], "files": ["usc_franka/video_000009.mp4", "usc_franka/video_000010.mp4"], "task_norm": "Put the yellow plate in the sink", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the yellow plate in the sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/169dd0a1a099a820b606574e.mp4"}, {"type": "video", "uri": "media/robometer_ood/a3ff5fb15da02c4731ec5e03.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000012-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["94af8ff4-0f77-48fd-9a83-cc35fd060e66", "f86a35b5-ab99-4a53-baed-78f83995ef74"], "files": ["usc_franka/video_000006.mp4", "usc_franka/video_000012.mp4"], "task_norm": "Put the yellow plate in the sink", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the yellow plate in the sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a3ff5fb15da02c4731ec5e03.mp4"}, {"type": "video", "uri": "media/robometer_ood/1d7e7efe744ea7801218e8e7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000012-vs-video_000008/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["f86a35b5-ab99-4a53-baed-78f83995ef74", "6f1c7142-25fe-4fd3-b6ff-7b1488a2aec1"], "files": ["usc_franka/video_000012.mp4", "usc_franka/video_000008.mp4"], "task_norm": "Put the yellow plate in the sink", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put banana on plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5566cdb6745c1ef5dfb3ffa9.mp4"}, {"type": "video", "uri": "media/robometer_ood/0b460b0a2cb424ffe89d3edf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000014-vs-video_000011/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["3e2c66e0-e1c0-4455-b51b-46af02d2cd03", "c006a720-9007-4088-bd91-7b570c1bb238"], "files": ["usc_franka/video_000014.mp4", "usc_franka/video_000011.mp4"], "task_norm": "Put banana on plate", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Put banana on plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7ea767442e6419df185109b3.mp4"}, {"type": "video", "uri": "media/robometer_ood/0c3ce8d98b888d0a83da85e0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000015-vs-video_000013/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["5966d747-adeb-4fd5-ae45-778efc7b0e08", "a2dae243-c742-4d25-a09d-d18b6c0baebe"], "files": ["usc_franka/video_000015.mp4", "usc_franka/video_000013.mp4"], "task_norm": "Put banana on plate", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Put banana on plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f3a9410d768e4e46f6329ead.mp4"}, {"type": "video", "uri": "media/robometer_ood/0c3ce8d98b888d0a83da85e0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000016-vs-video_000013/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["5ff4a784-0bf9-42cc-9e69-66548ca8b1e8", "a2dae243-c742-4d25-a09d-d18b6c0baebe"], "files": ["usc_franka/video_000016.mp4", "usc_franka/video_000013.mp4"], "task_norm": "Put banana on plate", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Put banana on plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5566cdb6745c1ef5dfb3ffa9.mp4"}, {"type": "video", "uri": "media/robometer_ood/f3a9410d768e4e46f6329ead.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000016-vs-video_000014/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["3e2c66e0-e1c0-4455-b51b-46af02d2cd03", "5ff4a784-0bf9-42cc-9e69-66548ca8b1e8"], "files": ["usc_franka/video_000014.mp4", "usc_franka/video_000016.mp4"], "task_norm": "Put banana on plate", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Put banana on plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0effdd669890be15a200aedf.mp4"}, {"type": "video", "uri": "media/robometer_ood/0b460b0a2cb424ffe89d3edf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000017-vs-video_000011/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["115e12f1-8e7a-429c-bc7a-6e4086649d5d", "c006a720-9007-4088-bd91-7b570c1bb238"], "files": ["usc_franka/video_000017.mp4", "usc_franka/video_000011.mp4"], "task_norm": "Put banana on plate", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Put banana on plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7ea767442e6419df185109b3.mp4"}, {"type": "video", "uri": "media/robometer_ood/0effdd669890be15a200aedf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000017-vs-video_000015/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["5966d747-adeb-4fd5-ae45-778efc7b0e08", "115e12f1-8e7a-429c-bc7a-6e4086649d5d"], "files": ["usc_franka/video_000015.mp4", "usc_franka/video_000017.mp4"], "task_norm": "Put banana on plate", "video_info": [{"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Stack the red cube on the gray cube", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/be5876f7afa5475d77490e3e.mp4"}, {"type": "video", "uri": "media/robometer_ood/f306027f4950e3f33f8108a0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000020-vs-video_000018/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["1419e514-1b2f-4c3a-a090-ffce17c5f95e", "4804b50b-9efa-4844-8e4d-533bd403f71e"], "files": ["usc_franka/video_000020.mp4", "usc_franka/video_000018.mp4"], "task_norm": "Stack the red cube on the gray cube", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the red cube on the gray cube", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f3e6d03f71cea17069f7cbc3.mp4"}, {"type": "video", "uri": "media/robometer_ood/3375e5c9332febf043d2623f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000021-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["0db11280-414a-4e11-81e1-2ee5e84c0262", "a4ed8e1b-48f5-4125-9972-c07674af249e"], "files": ["usc_franka/video_000021.mp4", "usc_franka/video_000019.mp4"], "task_norm": "Stack the red cube on the gray cube", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the red cube on the gray cube", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/580347ff7bc61f52a8906aab.mp4"}, {"type": "video", "uri": "media/robometer_ood/f306027f4950e3f33f8108a0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000022-vs-video_000018/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["b9eb4443-1d51-4376-822a-aa3492642d04", "4804b50b-9efa-4844-8e4d-533bd403f71e"], "files": ["usc_franka/video_000022.mp4", "usc_franka/video_000018.mp4"], "task_norm": "Stack the red cube on the gray cube", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the red cube on the gray cube", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/be5876f7afa5475d77490e3e.mp4"}, {"type": "video", "uri": "media/robometer_ood/580347ff7bc61f52a8906aab.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000022-vs-video_000020/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["1419e514-1b2f-4c3a-a090-ffce17c5f95e", "b9eb4443-1d51-4376-822a-aa3492642d04"], "files": ["usc_franka/video_000020.mp4", "usc_franka/video_000022.mp4"], "task_norm": "Stack the red cube on the gray cube", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the red cube on the gray cube", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e5048e74b351be631af8deec.mp4"}, {"type": "video", "uri": "media/robometer_ood/3375e5c9332febf043d2623f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000023-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["f8d7a3e1-c44f-42c8-836c-f650283551e9", "a4ed8e1b-48f5-4125-9972-c07674af249e"], "files": ["usc_franka/video_000023.mp4", "usc_franka/video_000019.mp4"], "task_norm": "Stack the red cube on the gray cube", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the red cube on the gray cube", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e5048e74b351be631af8deec.mp4"}, {"type": "video", "uri": "media/robometer_ood/f3e6d03f71cea17069f7cbc3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000023-vs-video_000021/robometer_traj_pref", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["f8d7a3e1-c44f-42c8-836c-f650283551e9", "0db11280-414a-4e11-81e1-2ee5e84c0262"], "files": ["usc_franka/video_000023.mp4", "usc_franka/video_000021.mp4"], "task_norm": "Stack the red cube on the gray cube", "video_info": [{"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/aadd61dadee3f6f1c3d46da1.mp4"}, {"type": "video", "uri": "media/robometer_ood/0956462da3a8deee366a0fbe.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000050-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["825b35e5-08a7-4f2d-9a57-f2d7a5d3457a", "37541aa9-a26c-4a0b-b415-7ce1851e45fc"], "files": ["usc_koch/video_000000.mp4", "usc_koch/video_000050.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0956462da3a8deee366a0fbe.mp4"}, {"type": "video", "uri": "media/robometer_ood/c7957a71cdff9469a1dabf41.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000050-vs-video_000093/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["37541aa9-a26c-4a0b-b415-7ce1851e45fc", "236dcff5-06b4-4893-8ceb-629ad1dc3e2b"], "files": ["usc_koch/video_000050.mp4", "usc_koch/video_000093.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/361ec073714e8a52a48dc6a9.mp4"}, {"type": "video", "uri": "media/robometer_ood/d319f56706d5f687cc3e13f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000051-vs-video_000003/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["cca1425f-b52a-43e3-b8c9-ab2f61ce9499", "997e4ad9-3b89-4254-86e8-7524534ac12e"], "files": ["usc_koch/video_000051.mp4", "usc_koch/video_000003.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f9512fa4efe1e3e4bab8ca11.mp4"}, {"type": "video", "uri": "media/robometer_ood/361ec073714e8a52a48dc6a9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000051-vs-video_000090/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["1d97116a-de3f-424c-94d2-0e2d149fdea0", "cca1425f-b52a-43e3-b8c9-ab2f61ce9499"], "files": ["usc_koch/video_000090.mp4", "usc_koch/video_000051.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7bc219471d970b672f96d637.mp4"}, {"type": "video", "uri": "media/robometer_ood/e539d046a6d3089883fc8523.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000052-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["1fa7f2e5-548a-410e-af82-1e2f11dbb0d5", "099d39c2-c632-4b33-bdb4-2ab2b870b7b9"], "files": ["usc_koch/video_000052.mp4", "usc_koch/video_000001.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7bc219471d970b672f96d637.mp4"}, {"type": "video", "uri": "media/robometer_ood/2e60e814ed70a5a801c0f377.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000052-vs-video_000091/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["1fa7f2e5-548a-410e-af82-1e2f11dbb0d5", "2a94cb79-36e6-4305-a0d4-5731b5b37a0f"], "files": ["usc_koch/video_000052.mp4", "usc_koch/video_000091.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/92d55fe186c736f4e935e923.mp4"}, {"type": "video", "uri": "media/robometer_ood/50a5c1674d4331d9ddc8a9b4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000053-vs-video_000004/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["5ed9b84b-e9f9-4557-ae59-e805128720f6", "dcaf97d1-1799-4110-9dc1-daeca882e461"], "files": ["usc_koch/video_000004.mp4", "usc_koch/video_000053.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/50a5c1674d4331d9ddc8a9b4.mp4"}, {"type": "video", "uri": "media/robometer_ood/e85774f4fe48bd9df7bda65f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000053-vs-video_000092/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["dcaf97d1-1799-4110-9dc1-daeca882e461", "8ba78da9-e03b-49be-a32c-35e2a3a19764"], "files": ["usc_koch/video_000053.mp4", "usc_koch/video_000092.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0a008df3bdb5751e6d1b3923.mp4"}, {"type": "video", "uri": "media/robometer_ood/fc64ee303784fc1ba340ad6e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000054-vs-video_000002/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["848b7b3f-c3cd-4814-a185-34216025cb90", "4136d3de-2fa3-449a-a9c4-d9a9206ded09"], "files": ["usc_koch/video_000002.mp4", "usc_koch/video_000054.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fc64ee303784fc1ba340ad6e.mp4"}, {"type": "video", "uri": "media/robometer_ood/c4e91d7a6e7e5b4a5b005190.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000054-vs-video_000094/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["4136d3de-2fa3-449a-a9c4-d9a9206ded09", "c87104bd-d3e8-4152-a7d5-fa62e7af576b"], "files": ["usc_koch/video_000054.mp4", "usc_koch/video_000094.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c131a61907d66872389aa607.mp4"}, {"type": "video", "uri": "media/robometer_ood/774de4523c7e0b2ab9c42dc7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000055-vs-video_000008/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["46a72685-0ed3-43d9-9154-88d8c4d90c7d", "9297e8f6-4440-40ac-8179-441b4d49c483"], "files": ["usc_koch/video_000055.mp4", "usc_koch/video_000008.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a1dd42fabb99a26aa24b6bd3.mp4"}, {"type": "video", "uri": "media/robometer_ood/c131a61907d66872389aa607.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000055-vs-video_000098/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["fc583831-0d7a-432c-b0ba-23d7d6aca68a", "46a72685-0ed3-43d9-9154-88d8c4d90c7d"], "files": ["usc_koch/video_000098.mp4", "usc_koch/video_000055.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2eca56fc9e349a0e5e0af584.mp4"}, {"type": "video", "uri": "media/robometer_ood/21075a9cf308415285685c4f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000056-vs-video_000009/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["1271eb80-7909-42e8-8a04-52c839be6a44", "700cbb25-27e3-4b65-ad81-502204b34071"], "files": ["usc_koch/video_000056.mp4", "usc_koch/video_000009.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5286d42cab3e7c4dcaf465ca.mp4"}, {"type": "video", "uri": "media/robometer_ood/2eca56fc9e349a0e5e0af584.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000056-vs-video_000097/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["fa3fc20e-130d-4035-b2f4-4d87c955ba45", "1271eb80-7909-42e8-8a04-52c839be6a44"], "files": ["usc_koch/video_000097.mp4", "usc_koch/video_000056.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/08472d3702906321d46f3728.mp4"}, {"type": "video", "uri": "media/robometer_ood/00f490892621e59be43fed0b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000057-vs-video_000005/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["7af3dd22-3db0-4426-938b-47ec12c1e667", "24f819f4-0a30-4e20-9573-d17e5ef2fc11"], "files": ["usc_koch/video_000057.mp4", "usc_koch/video_000005.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 21}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/be94e673ac6857989f892ed1.mp4"}, {"type": "video", "uri": "media/robometer_ood/08472d3702906321d46f3728.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000057-vs-video_000099/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["58c08402-79c6-43bb-b699-e2751751f21d", "7af3dd22-3db0-4426-938b-47ec12c1e667"], "files": ["usc_koch/video_000099.mp4", "usc_koch/video_000057.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/36547d6e5fe1b18d5eb9b507.mp4"}, {"type": "video", "uri": "media/robometer_ood/6256c96f60195bc0d17f5abf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000058-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["cce54bfa-775a-446e-bad7-4ebd0d4e6428", "0d368af7-daf1-411e-89d1-132d17ec73f5"], "files": ["usc_koch/video_000058.mp4", "usc_koch/video_000006.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b823a055cc8818aa6f9b9d63.mp4"}, {"type": "video", "uri": "media/robometer_ood/36547d6e5fe1b18d5eb9b507.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000058-vs-video_000095/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["129a48e3-fa70-45e8-ab64-09fd2084522c", "cce54bfa-775a-446e-bad7-4ebd0d4e6428"], "files": ["usc_koch/video_000095.mp4", "usc_koch/video_000058.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3427136fafbefdddd8748b01.mp4"}, {"type": "video", "uri": "media/robometer_ood/27a83fae7cee812682e800ae.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000059-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["1821b533-7108-4abf-9faa-bd68a0698ee4", "7e6acdf7-1393-4cd0-8de1-cc4fe7590380"], "files": ["usc_koch/video_000007.mp4", "usc_koch/video_000059.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1bc2b261e6351db0ede5386f.mp4"}, {"type": "video", "uri": "media/robometer_ood/27a83fae7cee812682e800ae.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000059-vs-video_000096/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["82d8a363-0055-4e4a-aca9-5a5fdfc8c0db", "7e6acdf7-1393-4cd0-8de1-cc4fe7590380"], "files": ["usc_koch/video_000096.mp4", "usc_koch/video_000059.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4bfc2ef32f3f924e0e68d891.mp4"}, {"type": "video", "uri": "media/robometer_ood/1cee21b2ba5c24d44d609971.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000060-vs-video_000011/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["9c1ce8e5-06eb-48ae-9a2a-fedf04db52d6", "b1224526-a8c1-4bc8-ba54-f80e5f6aa33c"], "files": ["usc_koch/video_000011.mp4", "usc_koch/video_000060.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/47de2518412bd11091c1f5e9.mp4"}, {"type": "video", "uri": "media/robometer_ood/1cee21b2ba5c24d44d609971.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000060-vs-video_000104/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["f377959d-9026-4ff0-b2d1-3e1f152f2b9a", "b1224526-a8c1-4bc8-ba54-f80e5f6aa33c"], "files": ["usc_koch/video_000104.mp4", "usc_koch/video_000060.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b867953b8dab43632565130e.mp4"}, {"type": "video", "uri": "media/robometer_ood/c7672da87e1edbbcbcfb9aad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000061-vs-video_000012/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["0687805b-46f9-40d0-b637-a58fe12db327", "bade4079-d263-4f66-afae-22c87b06a3e9"], "files": ["usc_koch/video_000012.mp4", "usc_koch/video_000061.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2bc97fa677d46be47d983292.mp4"}, {"type": "video", "uri": "media/robometer_ood/c7672da87e1edbbcbcfb9aad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000061-vs-video_000102/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["140a24d6-fd5c-453a-a8f2-609d0dc483c4", "bade4079-d263-4f66-afae-22c87b06a3e9"], "files": ["usc_koch/video_000102.mp4", "usc_koch/video_000061.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/40eaeb6ccd25447efe5e4926.mp4"}, {"type": "video", "uri": "media/robometer_ood/2076fad185f6cf6aa8944a69.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000062-vs-video_000013/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["cbc83d7d-c80b-46ad-9d78-5991291a78a2", "a4c50b45-4262-491b-9333-aa2628454c58"], "files": ["usc_koch/video_000013.mp4", "usc_koch/video_000062.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/724ea1ea8c005a8795d6a53a.mp4"}, {"type": "video", "uri": "media/robometer_ood/2076fad185f6cf6aa8944a69.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000062-vs-video_000101/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["e84950b9-6bb4-4e4b-994c-1fe66ea1205e", "a4c50b45-4262-491b-9333-aa2628454c58"], "files": ["usc_koch/video_000101.mp4", "usc_koch/video_000062.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fdfba1a57177e17896ce7044.mp4"}, {"type": "video", "uri": "media/robometer_ood/8393e9044843eb9cf143b93f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000063-vs-video_000014/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["1b743da9-8a11-4058-8dff-01f92e774ca9", "a65d04c6-cef3-40b8-97a1-f811dfabd609"], "files": ["usc_koch/video_000063.mp4", "usc_koch/video_000014.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fdfba1a57177e17896ce7044.mp4"}, {"type": "video", "uri": "media/robometer_ood/b66458b490ab70d3d0f4e71d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000063-vs-video_000103/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["1b743da9-8a11-4058-8dff-01f92e774ca9", "d74dbbac-db5a-48ee-a3b3-b29441d53ac4"], "files": ["usc_koch/video_000063.mp4", "usc_koch/video_000103.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0ea8c376475214ad385cad07.mp4"}, {"type": "video", "uri": "media/robometer_ood/68ca2c772fc944b62d2042bd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000064-vs-video_000010/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["474ce398-376a-4781-a953-c15b4cc9c865", "7d6947fd-1a74-4d77-852e-c2c9b465b95a"], "files": ["usc_koch/video_000010.mp4", "usc_koch/video_000064.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/68ca2c772fc944b62d2042bd.mp4"}, {"type": "video", "uri": "media/robometer_ood/45bbbebdeeb7957ac4146b02.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000064-vs-video_000100/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["7d6947fd-1a74-4d77-852e-c2c9b465b95a", "eeecf1c8-fd50-4cfe-9b45-91f4b8a25f94"], "files": ["usc_koch/video_000064.mp4", "usc_koch/video_000100.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f8d68ad43f3091cfc95ae819.mp4"}, {"type": "video", "uri": "media/robometer_ood/e09318bd6694a36675ea7bc3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000065-vs-video_000018/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["a3571be2-f4a8-4ea3-8f54-1fa4733e6e73", "1407d3b3-1d88-4df2-b8de-8cfbe0c5f314"], "files": ["usc_koch/video_000065.mp4", "usc_koch/video_000018.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f8d68ad43f3091cfc95ae819.mp4"}, {"type": "video", "uri": "media/robometer_ood/e91b6f4f6d9c3b4837f77d3f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000065-vs-video_000108/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["a3571be2-f4a8-4ea3-8f54-1fa4733e6e73", "2ac63e69-46a7-42e0-967b-c1a067ef406f"], "files": ["usc_koch/video_000065.mp4", "usc_koch/video_000108.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2ef32aba1377a43a294e4d3d.mp4"}, {"type": "video", "uri": "media/robometer_ood/9e5cbe69b8763b900b1db33e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000066-vs-video_000017/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["5a357cb6-7e5b-4dae-971f-c8a037611468", "e517076b-2ed5-4da3-89d0-4a5b2e1d84fa"], "files": ["usc_koch/video_000017.mp4", "usc_koch/video_000066.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1f9982263f99f0fa32ccb47c.mp4"}, {"type": "video", "uri": "media/robometer_ood/9e5cbe69b8763b900b1db33e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000066-vs-video_000107/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["6b155c2c-6297-410f-bc98-c4d2c44575cc", "e517076b-2ed5-4da3-89d0-4a5b2e1d84fa"], "files": ["usc_koch/video_000107.mp4", "usc_koch/video_000066.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2375bfa823288c496727b5aa.mp4"}, {"type": "video", "uri": "media/robometer_ood/a4ab87bcccb9e7311952bc75.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000067-vs-video_000015/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["f83e7747-d87d-450f-9bda-c943c0a40717", "82e7242b-bfcc-416e-be0c-6e46ee0becb5"], "files": ["usc_koch/video_000067.mp4", "usc_koch/video_000015.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2375bfa823288c496727b5aa.mp4"}, {"type": "video", "uri": "media/robometer_ood/0259845d2625ded9a663a950.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000067-vs-video_000106/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["f83e7747-d87d-450f-9bda-c943c0a40717", "8a59e8fd-4ce3-44bd-9654-48c133e066c9"], "files": ["usc_koch/video_000067.mp4", "usc_koch/video_000106.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f7dfa774f8f50e78bd927355.mp4"}, {"type": "video", "uri": "media/robometer_ood/eec326ce75370b88d0b5cad8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000068-vs-video_000016/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["d26a7326-12bd-444e-a621-ec8736184397", "661419ab-d66a-4025-8b88-ed6891cc58a9"], "files": ["usc_koch/video_000016.mp4", "usc_koch/video_000068.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6321a136ff37e46461bd6d5d.mp4"}, {"type": "video", "uri": "media/robometer_ood/eec326ce75370b88d0b5cad8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000068-vs-video_000105/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["6a79410d-61dd-4f13-a9a0-6a358a5bef6e", "661419ab-d66a-4025-8b88-ed6891cc58a9"], "files": ["usc_koch/video_000105.mp4", "usc_koch/video_000068.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c39a245e61b7512fcbd1d0b0.mp4"}, {"type": "video", "uri": "media/robometer_ood/2659bf1b87a3ad112dc8cff4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000069-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["c3589188-8b59-462e-ba1e-23183b5608da", "e362ec6b-f1d9-4a00-9a4a-36e783f97dd2"], "files": ["usc_koch/video_000069.mp4", "usc_koch/video_000019.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c39a245e61b7512fcbd1d0b0.mp4"}, {"type": "video", "uri": "media/robometer_ood/9a661a6704b2f41e4b377ef9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000069-vs-video_000109/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["c3589188-8b59-462e-ba1e-23183b5608da", "8d00a2a8-b18d-461d-b1bc-89955e37b8e0"], "files": ["usc_koch/video_000069.mp4", "usc_koch/video_000109.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/012fe793785bd2f01425fac5.mp4"}, {"type": "video", "uri": "media/robometer_ood/68f9de7f8c87059299c0d928.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000070-vs-video_000020/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["0f6a53ad-2490-4687-a2d6-6478ad374733", "06dd6133-b5d2-4a67-a489-082cf4b4a188"], "files": ["usc_koch/video_000070.mp4", "usc_koch/video_000020.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/59897913c47473f949ae4377.mp4"}, {"type": "video", "uri": "media/robometer_ood/012fe793785bd2f01425fac5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000070-vs-video_000112/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["ae450648-32ca-4bcb-8bc8-567ce66a3853", "0f6a53ad-2490-4687-a2d6-6478ad374733"], "files": ["usc_koch/video_000112.mp4", "usc_koch/video_000070.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/acd01ad9d0ee4734ecc00867.mp4"}, {"type": "video", "uri": "media/robometer_ood/e632789105d68aecbe10c9fd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000071-vs-video_000022/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["cf7f8d89-b361-47fe-a806-126bf4d62406", "a21e3d03-514c-4074-9de6-808b45be72ac"], "files": ["usc_koch/video_000071.mp4", "usc_koch/video_000022.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/167bc096c215bf44b7bfef5a.mp4"}, {"type": "video", "uri": "media/robometer_ood/acd01ad9d0ee4734ecc00867.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000071-vs-video_000110/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["9a29e5fd-08c3-4fa4-9978-dee2471596b5", "cf7f8d89-b361-47fe-a806-126bf4d62406"], "files": ["usc_koch/video_000110.mp4", "usc_koch/video_000071.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bc4bc2581aa6503b3ef26e19.mp4"}, {"type": "video", "uri": "media/robometer_ood/8edd63615a9a2d341073dcfb.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000072-vs-video_000024/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["fbead96c-1271-4b3c-b77d-f48a21b2a0bd", "29f4a914-5bec-4cf9-b8e0-05fd611fc207"], "files": ["usc_koch/video_000024.mp4", "usc_koch/video_000072.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b6bda94cd5e4627c295ecb51.mp4"}, {"type": "video", "uri": "media/robometer_ood/8edd63615a9a2d341073dcfb.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000072-vs-video_000113/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["ae740acc-fe67-46e6-83bd-c4223580b082", "29f4a914-5bec-4cf9-b8e0-05fd611fc207"], "files": ["usc_koch/video_000113.mp4", "usc_koch/video_000072.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/368d5d07602b6b6913115f72.mp4"}, {"type": "video", "uri": "media/robometer_ood/cb053787c218021cda06a9f1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000073-vs-video_000021/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["7f68f4e1-949a-44fd-ace0-d2fc23cff087", "5d703190-0df1-4daa-9088-0a45852f696c"], "files": ["usc_koch/video_000021.mp4", "usc_koch/video_000073.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4c2ab2dbdd886128441275e0.mp4"}, {"type": "video", "uri": "media/robometer_ood/cb053787c218021cda06a9f1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000073-vs-video_000111/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["e81e07fb-0c38-4f68-b012-9041b4a6f26e", "5d703190-0df1-4daa-9088-0a45852f696c"], "files": ["usc_koch/video_000111.mp4", "usc_koch/video_000073.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a6d6f633615b3e9dc9852020.mp4"}, {"type": "video", "uri": "media/robometer_ood/d8216dd71cdfbc33221a275d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000074-vs-video_000023/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["1b1f62db-8d77-437e-875a-c2d622fe36bd", "0aed6592-49f7-48c0-96ad-dc7e9c7e7ad1"], "files": ["usc_koch/video_000074.mp4", "usc_koch/video_000023.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a6d6f633615b3e9dc9852020.mp4"}, {"type": "video", "uri": "media/robometer_ood/accf9dd040cc22ed76e22944.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000074-vs-video_000114/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["1b1f62db-8d77-437e-875a-c2d622fe36bd", "512f705a-a2d2-4091-80c5-30de2b76eccb"], "files": ["usc_koch/video_000074.mp4", "usc_koch/video_000114.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1b5497c6d672b1215ed8ab2e.mp4"}, {"type": "video", "uri": "media/robometer_ood/af1a51bb4815076f2f07fc48.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000075-vs-video_000029/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["8514edb2-eba2-47f0-bafe-8eba4e21581e", "9dbf386f-4f8f-478c-a1b7-c82d5824ecbc"], "files": ["usc_koch/video_000075.mp4", "usc_koch/video_000029.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1b5497c6d672b1215ed8ab2e.mp4"}, {"type": "video", "uri": "media/robometer_ood/6326a077d4e8346d28e00290.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000075-vs-video_000118/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["8514edb2-eba2-47f0-bafe-8eba4e21581e", "dc8b5c16-f87f-4d9c-b063-7e71fdaf1452"], "files": ["usc_koch/video_000075.mp4", "usc_koch/video_000118.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/754f9d217b3fb8d2794507bf.mp4"}, {"type": "video", "uri": "media/robometer_ood/67d3247305e19238750ac1fc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000076-vs-video_000028/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["135694d3-39d1-4e45-9949-8fd75fc635c5", "d7d5bb5c-84a3-4ca5-945a-d3178f9c0f34"], "files": ["usc_koch/video_000028.mp4", "usc_koch/video_000076.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0cd8bca873c2fb38f89cb47f.mp4"}, {"type": "video", "uri": "media/robometer_ood/67d3247305e19238750ac1fc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000076-vs-video_000119/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["7d05f4dc-b0f6-4d30-9d7b-60c653b08025", "d7d5bb5c-84a3-4ca5-945a-d3178f9c0f34"], "files": ["usc_koch/video_000119.mp4", "usc_koch/video_000076.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/18e7f9428a91d37fc762d3c0.mp4"}, {"type": "video", "uri": "media/robometer_ood/c8c6f3842918bfe48f7163f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000077-vs-video_000025/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["2ee340e3-bc00-4860-b6bf-b23b96fd0359", "678b2d21-3e27-41a4-88a6-0d8a39d58a5e"], "files": ["usc_koch/video_000025.mp4", "usc_koch/video_000077.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/86c328b70138a6dacff92a64.mp4"}, {"type": "video", "uri": "media/robometer_ood/c8c6f3842918bfe48f7163f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000077-vs-video_000115/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["f8ecad70-7f93-474c-afea-1107f604e78c", "678b2d21-3e27-41a4-88a6-0d8a39d58a5e"], "files": ["usc_koch/video_000115.mp4", "usc_koch/video_000077.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ed24f6714f3fba005ac574cc.mp4"}, {"type": "video", "uri": "media/robometer_ood/c8ee74ba64324c30210896e1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000078-vs-video_000026/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["e69006dd-9929-4778-9ac8-772778533dcf", "a9adf6ac-ad21-4ee8-93aa-f8cd609daa6a"], "files": ["usc_koch/video_000078.mp4", "usc_koch/video_000026.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f3f00ceb50f8a842700f2624.mp4"}, {"type": "video", "uri": "media/robometer_ood/ed24f6714f3fba005ac574cc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000078-vs-video_000116/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["3bb5bd1f-d594-4200-a7d2-d716b21b324b", "e69006dd-9929-4778-9ac8-772778533dcf"], "files": ["usc_koch/video_000116.mp4", "usc_koch/video_000078.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f60c76083e6a60ec289d09e9.mp4"}, {"type": "video", "uri": "media/robometer_ood/5f62084205c68a23f829e8b7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000079-vs-video_000027/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["98b482ba-5984-4891-b068-69207b5896ee", "4d324a24-2ef1-4db0-b336-cca23179abee"], "files": ["usc_koch/video_000027.mp4", "usc_koch/video_000079.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4ec741bc87e33fa9e82a7b0b.mp4"}, {"type": "video", "uri": "media/robometer_ood/5f62084205c68a23f829e8b7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000079-vs-video_000117/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["2b03b363-5be0-42f3-92ac-3d334aea1ec1", "4d324a24-2ef1-4db0-b336-cca23179abee"], "files": ["usc_koch/video_000117.mp4", "usc_koch/video_000079.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1d56a18fd02ea151842e85b0.mp4"}, {"type": "video", "uri": "media/robometer_ood/f3c3c04f402b7edee41a062c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000080-vs-video_000033/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["6b399eea-7a64-47ee-acb4-cce20dcea553", "eb59f095-c125-473d-b305-22462d7b66cb"], "files": ["usc_koch/video_000080.mp4", "usc_koch/video_000033.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e04f3e0cd9219fe1d7b8d53d.mp4"}, {"type": "video", "uri": "media/robometer_ood/1d56a18fd02ea151842e85b0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000080-vs-video_000120/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["111ab8f1-c0ec-4882-a69c-1b2338c11ecd", "6b399eea-7a64-47ee-acb4-cce20dcea553"], "files": ["usc_koch/video_000120.mp4", "usc_koch/video_000080.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0b2dc3f03ab62ae326a6ed90.mp4"}, {"type": "video", "uri": "media/robometer_ood/726df9d19ea580f8d96f924e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000081-vs-video_000034/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["504099e1-3595-4b81-b8ff-99bfc6e68b88", "45d64f1b-a29c-4319-9589-aeb65ed8a153"], "files": ["usc_koch/video_000034.mp4", "usc_koch/video_000081.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ab9037cf120516fedea066a4.mp4"}, {"type": "video", "uri": "media/robometer_ood/726df9d19ea580f8d96f924e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000081-vs-video_000122/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["8e744467-2262-4c77-9e7b-9a7db709093b", "45d64f1b-a29c-4319-9589-aeb65ed8a153"], "files": ["usc_koch/video_000122.mp4", "usc_koch/video_000081.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c10fa8f778c2cea6498fd21b.mp4"}, {"type": "video", "uri": "media/robometer_ood/ced8a313f5e0a7b70f773197.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000082-vs-video_000030/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["46cd47bb-f690-4937-96a3-124567a31bb3", "1fe53487-ca8b-403a-b429-36a8d1250447"], "files": ["usc_koch/video_000082.mp4", "usc_koch/video_000030.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c10fa8f778c2cea6498fd21b.mp4"}, {"type": "video", "uri": "media/robometer_ood/591f6644fb783037f2b1057d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000082-vs-video_000121/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["46cd47bb-f690-4937-96a3-124567a31bb3", "f376c39e-6911-418a-aeae-7fd4bd6ce73c"], "files": ["usc_koch/video_000082.mp4", "usc_koch/video_000121.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ea34be0b5fdbf81915d1067d.mp4"}, {"type": "video", "uri": "media/robometer_ood/0d18b1acd5e5e65efbb3a687.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000083-vs-video_000031/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["b7071bb3-99e1-4af1-82f6-c19bbea69911", "de2eaf57-eb9d-4b9c-a02a-9042cfc89c3f"], "files": ["usc_koch/video_000031.mp4", "usc_koch/video_000083.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8347e52b013445ee4bc021f5.mp4"}, {"type": "video", "uri": "media/robometer_ood/0d18b1acd5e5e65efbb3a687.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000083-vs-video_000123/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["4f03c3d7-159a-42d5-8b66-2ea6063d6a40", "de2eaf57-eb9d-4b9c-a02a-9042cfc89c3f"], "files": ["usc_koch/video_000123.mp4", "usc_koch/video_000083.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/00c8dc0856e5e04ce2b7ba73.mp4"}, {"type": "video", "uri": "media/robometer_ood/fac15c4126282488f6bad150.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000084-vs-video_000032/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["dd8c85b1-a78d-4427-89e7-13389a2452af", "0e417347-7f7d-41ff-89d1-000a52ed164c"], "files": ["usc_koch/video_000084.mp4", "usc_koch/video_000032.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0922555c13a101c6690ec31b.mp4"}, {"type": "video", "uri": "media/robometer_ood/00c8dc0856e5e04ce2b7ba73.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000084-vs-video_000124/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["7d38eaeb-0dec-4476-af1a-80a4470ba08a", "dd8c85b1-a78d-4427-89e7-13389a2452af"], "files": ["usc_koch/video_000124.mp4", "usc_koch/video_000084.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/481022d159e4dcc05f126069.mp4"}, {"type": "video", "uri": "media/robometer_ood/50d02b44801a93d4e8729e87.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000085-vs-video_000039/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["7b5064ae-99eb-40ac-92e1-461eaf68f42f", "84484d2d-5cfa-4e80-9e4c-fd9657fb5bb1"], "files": ["usc_koch/video_000085.mp4", "usc_koch/video_000039.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/481022d159e4dcc05f126069.mp4"}, {"type": "video", "uri": "media/robometer_ood/8026ad252bd40577d04777c2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000085-vs-video_000127/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["7b5064ae-99eb-40ac-92e1-461eaf68f42f", "db81fda6-d347-4235-9a47-feed1655eed5"], "files": ["usc_koch/video_000085.mp4", "usc_koch/video_000127.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/417c6cc6f78ee2802d6357cd.mp4"}, {"type": "video", "uri": "media/robometer_ood/57a3da9999c6a617d40dd6a5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000086-vs-video_000038/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["83f1b266-7b3a-4104-8122-2f9e99ab60e2", "4737d874-045b-4596-aa0d-8841baca7581"], "files": ["usc_koch/video_000038.mp4", "usc_koch/video_000086.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1850b7d77faa445142490f74.mp4"}, {"type": "video", "uri": "media/robometer_ood/57a3da9999c6a617d40dd6a5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000086-vs-video_000129/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["0bb844f7-0ba8-4b42-afdf-b6f591db011d", "4737d874-045b-4596-aa0d-8841baca7581"], "files": ["usc_koch/video_000129.mp4", "usc_koch/video_000086.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9b929b7edf15ea22f6fdd837.mp4"}, {"type": "video", "uri": "media/robometer_ood/0d04647b916686027a7e9f26.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000087-vs-video_000036/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["a80e0d43-99e8-432f-ac92-865377371225", "e370fed8-781f-4fb9-87f0-72f2cd492876"], "files": ["usc_koch/video_000087.mp4", "usc_koch/video_000036.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5c1f2f905be2d6f08f3b79c1.mp4"}, {"type": "video", "uri": "media/robometer_ood/9b929b7edf15ea22f6fdd837.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000087-vs-video_000125/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["fa617e09-5d6a-4886-9168-b0847e47d4ac", "a80e0d43-99e8-432f-ac92-865377371225"], "files": ["usc_koch/video_000125.mp4", "usc_koch/video_000087.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/623383348dff7417b38284d8.mp4"}, {"type": "video", "uri": "media/robometer_ood/73d5dcd8fe55da89715d5452.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000088-vs-video_000037/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["b979938e-e8ac-4342-b3df-308bf7cdc2d7", "a670ad76-0983-4b9a-b59a-6d4f76bb29fa"], "files": ["usc_koch/video_000037.mp4", "usc_koch/video_000088.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c981451aa92e5a7904e0d5a1.mp4"}, {"type": "video", "uri": "media/robometer_ood/73d5dcd8fe55da89715d5452.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000088-vs-video_000126/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["7cce72fc-b8e8-49ef-90cb-6e9d84231729", "a670ad76-0983-4b9a-b59a-6d4f76bb29fa"], "files": ["usc_koch/video_000126.mp4", "usc_koch/video_000088.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/358971561d8c777b17300e95.mp4"}, {"type": "video", "uri": "media/robometer_ood/2a8399ad3e9d13dbdb4beda5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000089-vs-video_000035/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["6789878d-b12a-4666-a897-329f07d794c8", "ea21ecd7-794b-48a3-a79b-983ed6d562d9"], "files": ["usc_koch/video_000089.mp4", "usc_koch/video_000035.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/358971561d8c777b17300e95.mp4"}, {"type": "video", "uri": "media/robometer_ood/1b185c28158a1b4f3bae4b3d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000089-vs-video_000128/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["6789878d-b12a-4666-a897-329f07d794c8", "75d0b3b2-7a09-48b0-825e-616374644dc9"], "files": ["usc_koch/video_000089.mp4", "usc_koch/video_000128.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f9512fa4efe1e3e4bab8ca11.mp4"}, {"type": "video", "uri": "media/robometer_ood/aadd61dadee3f6f1c3d46da1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000090-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["1d97116a-de3f-424c-94d2-0e2d149fdea0", "825b35e5-08a7-4f2d-9a57-f2d7a5d3457a"], "files": ["usc_koch/video_000090.mp4", "usc_koch/video_000000.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2e60e814ed70a5a801c0f377.mp4"}, {"type": "video", "uri": "media/robometer_ood/0a008df3bdb5751e6d1b3923.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000091-vs-video_000002/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["2a94cb79-36e6-4305-a0d4-5731b5b37a0f", "848b7b3f-c3cd-4814-a185-34216025cb90"], "files": ["usc_koch/video_000091.mp4", "usc_koch/video_000002.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e85774f4fe48bd9df7bda65f.mp4"}, {"type": "video", "uri": "media/robometer_ood/d319f56706d5f687cc3e13f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000092-vs-video_000003/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["8ba78da9-e03b-49be-a32c-35e2a3a19764", "997e4ad9-3b89-4254-86e8-7524534ac12e"], "files": ["usc_koch/video_000092.mp4", "usc_koch/video_000003.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e539d046a6d3089883fc8523.mp4"}, {"type": "video", "uri": "media/robometer_ood/c7957a71cdff9469a1dabf41.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000093-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["099d39c2-c632-4b33-bdb4-2ab2b870b7b9", "236dcff5-06b4-4893-8ceb-629ad1dc3e2b"], "files": ["usc_koch/video_000001.mp4", "usc_koch/video_000093.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from left to right", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/92d55fe186c736f4e935e923.mp4"}, {"type": "video", "uri": "media/robometer_ood/c4e91d7a6e7e5b4a5b005190.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000094-vs-video_000004/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["5ed9b84b-e9f9-4557-ae59-e805128720f6", "c87104bd-d3e8-4152-a7d5-fa62e7af576b"], "files": ["usc_koch/video_000004.mp4", "usc_koch/video_000094.mp4"], "task_norm": "Move the orange cup from left to right", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b823a055cc8818aa6f9b9d63.mp4"}, {"type": "video", "uri": "media/robometer_ood/21075a9cf308415285685c4f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000095-vs-video_000009/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["129a48e3-fa70-45e8-ab64-09fd2084522c", "700cbb25-27e3-4b65-ad81-502204b34071"], "files": ["usc_koch/video_000095.mp4", "usc_koch/video_000009.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1bc2b261e6351db0ede5386f.mp4"}, {"type": "video", "uri": "media/robometer_ood/774de4523c7e0b2ab9c42dc7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000096-vs-video_000008/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["82d8a363-0055-4e4a-aca9-5a5fdfc8c0db", "9297e8f6-4440-40ac-8179-441b4d49c483"], "files": ["usc_koch/video_000096.mp4", "usc_koch/video_000008.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/00f490892621e59be43fed0b.mp4"}, {"type": "video", "uri": "media/robometer_ood/5286d42cab3e7c4dcaf465ca.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000097-vs-video_000005/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["24f819f4-0a30-4e20-9573-d17e5ef2fc11", "fa3fc20e-130d-4035-b2f4-4d87c955ba45"], "files": ["usc_koch/video_000005.mp4", "usc_koch/video_000097.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 21}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3427136fafbefdddd8748b01.mp4"}, {"type": "video", "uri": "media/robometer_ood/a1dd42fabb99a26aa24b6bd3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000098-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["1821b533-7108-4abf-9faa-bd68a0698ee4", "fc583831-0d7a-432c-b0ba-23d7d6aca68a"], "files": ["usc_koch/video_000007.mp4", "usc_koch/video_000098.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Move the orange cup from right to left", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/be94e673ac6857989f892ed1.mp4"}, {"type": "video", "uri": "media/robometer_ood/6256c96f60195bc0d17f5abf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000099-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["58c08402-79c6-43bb-b699-e2751751f21d", "0d368af7-daf1-411e-89d1-132d17ec73f5"], "files": ["usc_koch/video_000099.mp4", "usc_koch/video_000006.mp4"], "task_norm": "Move the orange cup from right to left", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/45bbbebdeeb7957ac4146b02.mp4"}, {"type": "video", "uri": "media/robometer_ood/40eaeb6ccd25447efe5e4926.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000100-vs-video_000013/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["eeecf1c8-fd50-4cfe-9b45-91f4b8a25f94", "cbc83d7d-c80b-46ad-9d78-5991291a78a2"], "files": ["usc_koch/video_000100.mp4", "usc_koch/video_000013.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/724ea1ea8c005a8795d6a53a.mp4"}, {"type": "video", "uri": "media/robometer_ood/0ea8c376475214ad385cad07.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000101-vs-video_000010/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["e84950b9-6bb4-4e4b-994c-1fe66ea1205e", "474ce398-376a-4781-a953-c15b4cc9c865"], "files": ["usc_koch/video_000101.mp4", "usc_koch/video_000010.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2bc97fa677d46be47d983292.mp4"}, {"type": "video", "uri": "media/robometer_ood/4bfc2ef32f3f924e0e68d891.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000102-vs-video_000011/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["140a24d6-fd5c-453a-a8f2-609d0dc483c4", "9c1ce8e5-06eb-48ae-9a2a-fedf04db52d6"], "files": ["usc_koch/video_000102.mp4", "usc_koch/video_000011.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8393e9044843eb9cf143b93f.mp4"}, {"type": "video", "uri": "media/robometer_ood/b66458b490ab70d3d0f4e71d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000103-vs-video_000014/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["a65d04c6-cef3-40b8-97a1-f811dfabd609", "d74dbbac-db5a-48ee-a3b3-b29441d53ac4"], "files": ["usc_koch/video_000014.mp4", "usc_koch/video_000103.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the blue trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b867953b8dab43632565130e.mp4"}, {"type": "video", "uri": "media/robometer_ood/47de2518412bd11091c1f5e9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000104-vs-video_000012/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["0687805b-46f9-40d0-b637-a58fe12db327", "f377959d-9026-4ff0-b2d1-3e1f152f2b9a"], "files": ["usc_koch/video_000012.mp4", "usc_koch/video_000104.mp4"], "task_norm": "Open the blue trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2659bf1b87a3ad112dc8cff4.mp4"}, {"type": "video", "uri": "media/robometer_ood/6321a136ff37e46461bd6d5d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000105-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["e362ec6b-f1d9-4a00-9a4a-36e783f97dd2", "6a79410d-61dd-4f13-a9a0-6a358a5bef6e"], "files": ["usc_koch/video_000019.mp4", "usc_koch/video_000105.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2ef32aba1377a43a294e4d3d.mp4"}, {"type": "video", "uri": "media/robometer_ood/0259845d2625ded9a663a950.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000106-vs-video_000017/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["5a357cb6-7e5b-4dae-971f-c8a037611468", "8a59e8fd-4ce3-44bd-9654-48c133e066c9"], "files": ["usc_koch/video_000017.mp4", "usc_koch/video_000106.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1f9982263f99f0fa32ccb47c.mp4"}, {"type": "video", "uri": "media/robometer_ood/a4ab87bcccb9e7311952bc75.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000107-vs-video_000015/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["6b155c2c-6297-410f-bc98-c4d2c44575cc", "82e7242b-bfcc-416e-be0c-6e46ee0becb5"], "files": ["usc_koch/video_000107.mp4", "usc_koch/video_000015.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f7dfa774f8f50e78bd927355.mp4"}, {"type": "video", "uri": "media/robometer_ood/e91b6f4f6d9c3b4837f77d3f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000108-vs-video_000016/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["d26a7326-12bd-444e-a621-ec8736184397", "2ac63e69-46a7-42e0-967b-c1a067ef406f"], "files": ["usc_koch/video_000016.mp4", "usc_koch/video_000108.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the green trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9a661a6704b2f41e4b377ef9.mp4"}, {"type": "video", "uri": "media/robometer_ood/e09318bd6694a36675ea7bc3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000109-vs-video_000018/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["8d00a2a8-b18d-461d-b1bc-89955e37b8e0", "1407d3b3-1d88-4df2-b8de-8cfbe0c5f314"], "files": ["usc_koch/video_000109.mp4", "usc_koch/video_000018.mp4"], "task_norm": "Open the green trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e632789105d68aecbe10c9fd.mp4"}, {"type": "video", "uri": "media/robometer_ood/167bc096c215bf44b7bfef5a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000110-vs-video_000022/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["a21e3d03-514c-4074-9de6-808b45be72ac", "9a29e5fd-08c3-4fa4-9978-dee2471596b5"], "files": ["usc_koch/video_000022.mp4", "usc_koch/video_000110.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/68f9de7f8c87059299c0d928.mp4"}, {"type": "video", "uri": "media/robometer_ood/4c2ab2dbdd886128441275e0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000111-vs-video_000020/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["06dd6133-b5d2-4a67-a489-082cf4b4a188", "e81e07fb-0c38-4f68-b012-9041b4a6f26e"], "files": ["usc_koch/video_000020.mp4", "usc_koch/video_000111.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d8216dd71cdfbc33221a275d.mp4"}, {"type": "video", "uri": "media/robometer_ood/59897913c47473f949ae4377.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000112-vs-video_000023/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["0aed6592-49f7-48c0-96ad-dc7e9c7e7ad1", "ae450648-32ca-4bcb-8bc8-567ce66a3853"], "files": ["usc_koch/video_000023.mp4", "usc_koch/video_000112.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b6bda94cd5e4627c295ecb51.mp4"}, {"type": "video", "uri": "media/robometer_ood/368d5d07602b6b6913115f72.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000113-vs-video_000021/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["ae740acc-fe67-46e6-83bd-c4223580b082", "7f68f4e1-949a-44fd-ace0-d2fc23cff087"], "files": ["usc_koch/video_000113.mp4", "usc_koch/video_000021.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Open the red trash bin", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bc4bc2581aa6503b3ef26e19.mp4"}, {"type": "video", "uri": "media/robometer_ood/accf9dd040cc22ed76e22944.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000114-vs-video_000024/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["fbead96c-1271-4b3c-b77d-f48a21b2a0bd", "512f705a-a2d2-4091-80c5-30de2b76eccb"], "files": ["usc_koch/video_000024.mp4", "usc_koch/video_000114.mp4"], "task_norm": "Open the red trash bin", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/af1a51bb4815076f2f07fc48.mp4"}, {"type": "video", "uri": "media/robometer_ood/86c328b70138a6dacff92a64.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000115-vs-video_000029/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["9dbf386f-4f8f-478c-a1b7-c82d5824ecbc", "f8ecad70-7f93-474c-afea-1107f604e78c"], "files": ["usc_koch/video_000029.mp4", "usc_koch/video_000115.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f3f00ceb50f8a842700f2624.mp4"}, {"type": "video", "uri": "media/robometer_ood/18e7f9428a91d37fc762d3c0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000116-vs-video_000025/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["3bb5bd1f-d594-4200-a7d2-d716b21b324b", "2ee340e3-bc00-4860-b6bf-b23b96fd0359"], "files": ["usc_koch/video_000116.mp4", "usc_koch/video_000025.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4ec741bc87e33fa9e82a7b0b.mp4"}, {"type": "video", "uri": "media/robometer_ood/c8ee74ba64324c30210896e1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000117-vs-video_000026/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["2b03b363-5be0-42f3-92ac-3d334aea1ec1", "a9adf6ac-ad21-4ee8-93aa-f8cd609daa6a"], "files": ["usc_koch/video_000117.mp4", "usc_koch/video_000026.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6326a077d4e8346d28e00290.mp4"}, {"type": "video", "uri": "media/robometer_ood/f60c76083e6a60ec289d09e9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000118-vs-video_000027/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["dc8b5c16-f87f-4d9c-b063-7e71fdaf1452", "98b482ba-5984-4891-b068-69207b5896ee"], "files": ["usc_koch/video_000118.mp4", "usc_koch/video_000027.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/754f9d217b3fb8d2794507bf.mp4"}, {"type": "video", "uri": "media/robometer_ood/0cd8bca873c2fb38f89cb47f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000119-vs-video_000028/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["135694d3-39d1-4e45-9949-8fd75fc635c5", "7d05f4dc-b0f6-4d30-9d7b-60c653b08025"], "files": ["usc_koch/video_000028.mp4", "usc_koch/video_000119.mp4"], "task_norm": "Separate the purple and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e04f3e0cd9219fe1d7b8d53d.mp4"}, {"type": "video", "uri": "media/robometer_ood/0b2dc3f03ab62ae326a6ed90.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000120-vs-video_000034/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["111ab8f1-c0ec-4882-a69c-1b2338c11ecd", "504099e1-3595-4b81-b8ff-99bfc6e68b88"], "files": ["usc_koch/video_000120.mp4", "usc_koch/video_000034.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ced8a313f5e0a7b70f773197.mp4"}, {"type": "video", "uri": "media/robometer_ood/591f6644fb783037f2b1057d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000121-vs-video_000030/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["1fe53487-ca8b-403a-b429-36a8d1250447", "f376c39e-6911-418a-aeae-7fd4bd6ce73c"], "files": ["usc_koch/video_000030.mp4", "usc_koch/video_000121.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f3c3c04f402b7edee41a062c.mp4"}, {"type": "video", "uri": "media/robometer_ood/ab9037cf120516fedea066a4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000122-vs-video_000033/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["eb59f095-c125-473d-b305-22462d7b66cb", "8e744467-2262-4c77-9e7b-9a7db709093b"], "files": ["usc_koch/video_000033.mp4", "usc_koch/video_000122.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ea34be0b5fdbf81915d1067d.mp4"}, {"type": "video", "uri": "media/robometer_ood/8347e52b013445ee4bc021f5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000123-vs-video_000031/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["b7071bb3-99e1-4af1-82f6-c19bbea69911", "4f03c3d7-159a-42d5-8b66-2ea6063d6a40"], "files": ["usc_koch/video_000031.mp4", "usc_koch/video_000123.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the purple and red cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fac15c4126282488f6bad150.mp4"}, {"type": "video", "uri": "media/robometer_ood/0922555c13a101c6690ec31b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000124-vs-video_000032/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["0e417347-7f7d-41ff-89d1-000a52ed164c", "7d38eaeb-0dec-4476-af1a-80a4470ba08a"], "files": ["usc_koch/video_000032.mp4", "usc_koch/video_000124.mp4"], "task_norm": "Separate the purple and red cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0d04647b916686027a7e9f26.mp4"}, {"type": "video", "uri": "media/robometer_ood/5c1f2f905be2d6f08f3b79c1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000125-vs-video_000036/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["e370fed8-781f-4fb9-87f0-72f2cd492876", "fa617e09-5d6a-4886-9168-b0847e47d4ac"], "files": ["usc_koch/video_000036.mp4", "usc_koch/video_000125.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/50d02b44801a93d4e8729e87.mp4"}, {"type": "video", "uri": "media/robometer_ood/c981451aa92e5a7904e0d5a1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000126-vs-video_000039/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["84484d2d-5cfa-4e80-9e4c-fd9657fb5bb1", "7cce72fc-b8e8-49ef-90cb-6e9d84231729"], "files": ["usc_koch/video_000039.mp4", "usc_koch/video_000126.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8026ad252bd40577d04777c2.mp4"}, {"type": "video", "uri": "media/robometer_ood/623383348dff7417b38284d8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000127-vs-video_000037/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["db81fda6-d347-4235-9a47-feed1655eed5", "b979938e-e8ac-4342-b3df-308bf7cdc2d7"], "files": ["usc_koch/video_000127.mp4", "usc_koch/video_000037.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/417c6cc6f78ee2802d6357cd.mp4"}, {"type": "video", "uri": "media/robometer_ood/1b185c28158a1b4f3bae4b3d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000128-vs-video_000038/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["83f1b266-7b3a-4104-8122-2f9e99ab60e2", "75d0b3b2-7a09-48b0-825e-616374644dc9"], "files": ["usc_koch/video_000038.mp4", "usc_koch/video_000128.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Separate the red and orange cups", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2a8399ad3e9d13dbdb4beda5.mp4"}, {"type": "video", "uri": "media/robometer_ood/1850b7d77faa445142490f74.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000129-vs-video_000035/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["ea21ecd7-794b-48a3-a79b-983ed6d562d9", "0bb844f7-0ba8-4b42-afdf-b6f591db011d"], "files": ["usc_koch/video_000035.mp4", "usc_koch/video_000129.mp4"], "task_norm": "Separate the red and orange cups", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7658b39ed7aa43655df45a86.mp4"}, {"type": "video", "uri": "media/robometer_ood/da8abc6a5d43d6e56477f2ce.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000130-vs-video_000043/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["0e5fbba7-8352-4276-861d-e74dbd72a5bf", "8b16a94c-da02-41e3-855a-683319ac1337"], "files": ["usc_koch/video_000130.mp4", "usc_koch/video_000043.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/62afe872fcce093fc7306595.mp4"}, {"type": "video", "uri": "media/robometer_ood/208d25be797f2a55a02c36d1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000131-vs-video_000041/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["38c66aba-3e7f-4143-9d42-cafd96a81488", "c5cfe7d3-629b-49a8-9457-8edb3279e0d8"], "files": ["usc_koch/video_000041.mp4", "usc_koch/video_000131.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1dc173a57909407873fa700a.mp4"}, {"type": "video", "uri": "media/robometer_ood/b33b2335f57212d16a4ad7f3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000132-vs-video_000044/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["b2c8a625-3de1-4da2-a704-b50ca0cbd6af", "40ac1dcc-1c23-4ad0-bf18-450eab7b5349"], "files": ["usc_koch/video_000044.mp4", "usc_koch/video_000132.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/163aece0bb1caee904cf9ea3.mp4"}, {"type": "video", "uri": "media/robometer_ood/a11cd67a62a7f5badc81c611.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000133-vs-video_000040/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["9dc92f66-db65-4f9d-a33d-04c453a22397", "ae9e0f4c-57e4-4b1a-affe-127dca818414"], "files": ["usc_koch/video_000133.mp4", "usc_koch/video_000040.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8cf35b07270ea0cb3270c37b.mp4"}, {"type": "video", "uri": "media/robometer_ood/48d79577d07a6b32264f6ed3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000134-vs-video_000042/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["084abdd0-1fe9-47dc-a5d0-9e67af3bdc73", "f3e2bbe5-64df-4ccd-9251-8bdb0bf8e195"], "files": ["usc_koch/video_000134.mp4", "usc_koch/video_000042.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/44ea7b6a831070e82b75d6c7.mp4"}, {"type": "video", "uri": "media/robometer_ood/7fad72f2860e8be2327ba98a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000135-vs-video_000045/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["6f567925-802b-4f5c-a8a0-69c783df27a2", "d56da746-11bc-4c11-8b3d-0f8d4497e24f"], "files": ["usc_koch/video_000135.mp4", "usc_koch/video_000045.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cd4476e6aa7b84fd0290bc22.mp4"}, {"type": "video", "uri": "media/robometer_ood/dd6cd2bc2f71f46571c17d37.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000136-vs-video_000047/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["adad6f55-083d-4228-a163-4355d1e4a12c", "65aa1d59-fd81-4609-9b05-1f09bd66bb47"], "files": ["usc_koch/video_000136.mp4", "usc_koch/video_000047.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2c5fb4347f49cd8002161422.mp4"}, {"type": "video", "uri": "media/robometer_ood/753281f3616738be8c4daaac.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000137-vs-video_000046/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["aed484f3-5e85-49e1-bd49-7eadce0fdb29", "7bfa319b-6bbe-4f5c-85ce-f9d8edf804d6"], "files": ["usc_koch/video_000046.mp4", "usc_koch/video_000137.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/36a8909fda4e5e4b2828efb9.mp4"}, {"type": "video", "uri": "media/robometer_ood/bdbe81437a702f595995e68b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000138-vs-video_000048/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["bc84aa26-44df-4e58-a969-3faa19fbafc0", "ada56406-a1c2-4228-bfc7-0f2d6c877ed7"], "files": ["usc_koch/video_000138.mp4", "usc_koch/video_000048.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3a6378ae400717fab11db48c.mp4"}, {"type": "video", "uri": "media/robometer_ood/11cb5ad61fd868a3f01eb451.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000139-vs-video_000049/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["9ad79592-a48c-4f28-b4e8-5ea8835f3717", "45d8c456-0ae2-4a23-9d5a-f9a946806d5b"], "files": ["usc_koch/video_000139.mp4", "usc_koch/video_000049.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/48d79577d07a6b32264f6ed3.mp4"}, {"type": "video", "uri": "media/robometer_ood/bff83d49038831fd47cd5e8f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000140-vs-video_000042/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["f3e2bbe5-64df-4ccd-9251-8bdb0bf8e195", "40073104-3e83-46ad-9af1-0c3bcf2ca384"], "files": ["usc_koch/video_000042.mp4", "usc_koch/video_000140.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bff83d49038831fd47cd5e8f.mp4"}, {"type": "video", "uri": "media/robometer_ood/7658b39ed7aa43655df45a86.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000140-vs-video_000130/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["40073104-3e83-46ad-9af1-0c3bcf2ca384", "0e5fbba7-8352-4276-861d-e74dbd72a5bf"], "files": ["usc_koch/video_000140.mp4", "usc_koch/video_000130.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/602d3a7ee2de086472e14717.mp4"}, {"type": "video", "uri": "media/robometer_ood/da8abc6a5d43d6e56477f2ce.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000141-vs-video_000043/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["5cfc02b2-5402-44d7-9314-d490dfb48f10", "8b16a94c-da02-41e3-855a-683319ac1337"], "files": ["usc_koch/video_000141.mp4", "usc_koch/video_000043.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/602d3a7ee2de086472e14717.mp4"}, {"type": "video", "uri": "media/robometer_ood/163aece0bb1caee904cf9ea3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000141-vs-video_000133/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["5cfc02b2-5402-44d7-9314-d490dfb48f10", "9dc92f66-db65-4f9d-a33d-04c453a22397"], "files": ["usc_koch/video_000141.mp4", "usc_koch/video_000133.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/62afe872fcce093fc7306595.mp4"}, {"type": "video", "uri": "media/robometer_ood/af2d593574521dbbfc7a65b2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000142-vs-video_000041/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["38c66aba-3e7f-4143-9d42-cafd96a81488", "61ce528d-4829-4878-9647-0498d3e2b6b9"], "files": ["usc_koch/video_000041.mp4", "usc_koch/video_000142.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/af2d593574521dbbfc7a65b2.mp4"}, {"type": "video", "uri": "media/robometer_ood/b33b2335f57212d16a4ad7f3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000142-vs-video_000132/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["61ce528d-4829-4878-9647-0498d3e2b6b9", "40ac1dcc-1c23-4ad0-bf18-450eab7b5349"], "files": ["usc_koch/video_000142.mp4", "usc_koch/video_000132.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a11cd67a62a7f5badc81c611.mp4"}, {"type": "video", "uri": "media/robometer_ood/2289bea937981ca481a9f7a3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000143-vs-video_000040/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["ae9e0f4c-57e4-4b1a-affe-127dca818414", "414634c3-c744-42f6-a9b3-249aef0316fa"], "files": ["usc_koch/video_000040.mp4", "usc_koch/video_000143.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2289bea937981ca481a9f7a3.mp4"}, {"type": "video", "uri": "media/robometer_ood/8cf35b07270ea0cb3270c37b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000143-vs-video_000134/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["414634c3-c744-42f6-a9b3-249aef0316fa", "084abdd0-1fe9-47dc-a5d0-9e67af3bdc73"], "files": ["usc_koch/video_000143.mp4", "usc_koch/video_000134.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1dc173a57909407873fa700a.mp4"}, {"type": "video", "uri": "media/robometer_ood/fca8a4488a2a44c08df29ee9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000144-vs-video_000044/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["b2c8a625-3de1-4da2-a704-b50ca0cbd6af", "36925fed-6878-4520-beac-0490de9cc03c"], "files": ["usc_koch/video_000044.mp4", "usc_koch/video_000144.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/208d25be797f2a55a02c36d1.mp4"}, {"type": "video", "uri": "media/robometer_ood/fca8a4488a2a44c08df29ee9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000144-vs-video_000131/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["c5cfe7d3-629b-49a8-9457-8edb3279e0d8", "36925fed-6878-4520-beac-0490de9cc03c"], "files": ["usc_koch/video_000131.mp4", "usc_koch/video_000144.mp4"], "task_norm": "Throw the black marker away in the blue trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b1dcc009840151fb1ebd8596.mp4"}, {"type": "video", "uri": "media/robometer_ood/2c5fb4347f49cd8002161422.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000145-vs-video_000046/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["b1f1ebcd-adc3-42f8-9a3b-f393f473b269", "aed484f3-5e85-49e1-bd49-7eadce0fdb29"], "files": ["usc_koch/video_000145.mp4", "usc_koch/video_000046.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b1dcc009840151fb1ebd8596.mp4"}, {"type": "video", "uri": "media/robometer_ood/44ea7b6a831070e82b75d6c7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000145-vs-video_000135/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["b1f1ebcd-adc3-42f8-9a3b-f393f473b269", "6f567925-802b-4f5c-a8a0-69c783df27a2"], "files": ["usc_koch/video_000145.mp4", "usc_koch/video_000135.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2b061e39d8bd94c2898cd0c4.mp4"}, {"type": "video", "uri": "media/robometer_ood/11cb5ad61fd868a3f01eb451.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000146-vs-video_000049/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["f30457a0-5cb7-4412-b873-e63d93de6ad1", "45d8c456-0ae2-4a23-9d5a-f9a946806d5b"], "files": ["usc_koch/video_000146.mp4", "usc_koch/video_000049.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3a6378ae400717fab11db48c.mp4"}, {"type": "video", "uri": "media/robometer_ood/2b061e39d8bd94c2898cd0c4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000146-vs-video_000139/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["9ad79592-a48c-4f28-b4e8-5ea8835f3717", "f30457a0-5cb7-4412-b873-e63d93de6ad1"], "files": ["usc_koch/video_000139.mp4", "usc_koch/video_000146.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7e22bb51ee3ec28b63622914.mp4"}, {"type": "video", "uri": "media/robometer_ood/7fad72f2860e8be2327ba98a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000147-vs-video_000045/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["1705cd6f-552d-4250-8e49-0f708c092b50", "d56da746-11bc-4c11-8b3d-0f8d4497e24f"], "files": ["usc_koch/video_000147.mp4", "usc_koch/video_000045.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7e22bb51ee3ec28b63622914.mp4"}, {"type": "video", "uri": "media/robometer_ood/753281f3616738be8c4daaac.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000147-vs-video_000137/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["1705cd6f-552d-4250-8e49-0f708c092b50", "7bfa319b-6bbe-4f5c-85ce-f9d8edf804d6"], "files": ["usc_koch/video_000147.mp4", "usc_koch/video_000137.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/09309f435866a485908b8520.mp4"}, {"type": "video", "uri": "media/robometer_ood/dd6cd2bc2f71f46571c17d37.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000148-vs-video_000047/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["def91146-877c-4277-83c1-741029f542bf", "65aa1d59-fd81-4609-9b05-1f09bd66bb47"], "files": ["usc_koch/video_000148.mp4", "usc_koch/video_000047.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cd4476e6aa7b84fd0290bc22.mp4"}, {"type": "video", "uri": "media/robometer_ood/09309f435866a485908b8520.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000148-vs-video_000136/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["adad6f55-083d-4228-a163-4355d1e4a12c", "def91146-877c-4277-83c1-741029f542bf"], "files": ["usc_koch/video_000136.mp4", "usc_koch/video_000148.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bdbe81437a702f595995e68b.mp4"}, {"type": "video", "uri": "media/robometer_ood/a5cb584e04b19b0250b44e10.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000149-vs-video_000048/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["ada56406-a1c2-4228-bfc7-0f2d6c877ed7", "3faeca25-e362-4701-933e-631858a0d3b7"], "files": ["usc_koch/video_000048.mp4", "usc_koch/video_000149.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/36a8909fda4e5e4b2828efb9.mp4"}, {"type": "video", "uri": "media/robometer_ood/a5cb584e04b19b0250b44e10.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000149-vs-video_000138/robometer_traj_pref", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["bc84aa26-44df-4e58-a969-3faa19fbafc0", "3faeca25-e362-4701-933e-631858a0d3b7"], "files": ["usc_koch/video_000138.mp4", "usc_koch/video_000149.mp4"], "task_norm": "Throw the orange cup away in the red trash can", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 158, "role": "preference"}}
{"state": {"task": "Open the bottle", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f98edd1610c74695b0ef170a.mp4"}, {"type": "video", "uri": "media/robometer_ood/9a5b9064f94e277a4f6361de.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000001-vs-video_000002/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Open the bottle", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["bfad213d-ef98-41ea-9023-174e018287a0", "c243382e-cb06-42fb-a1ce-ec787a7da218"], "files": ["usc_trossen/video_000002.mp4", "usc_trossen/video_000001.mp4"], "task_norm": "Open the bottle", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Open the bottle", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1b5134c61cf5fe4fba1f7768.mp4"}, {"type": "video", "uri": "media/robometer_ood/51a9b9b331dc4f29467359c1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000003-vs-video_000004/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Open the bottle", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["cc2a5e47-6f33-4763-9f7c-ece094b42225", "da5b17da-f053-47f0-81d4-275cd1bd668b"], "files": ["usc_trossen/video_000003.mp4", "usc_trossen/video_000004.mp4"], "task_norm": "Open the bottle", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Open the red drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8a7cb858a552adc296e98610.mp4"}, {"type": "video", "uri": "media/robometer_ood/9e35c22104c1dd795fbf8e5e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000005-vs-video_000009/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["fcdd7b74-542e-455f-aca8-f6bd110144a8", "fab793f7-2b04-4425-8af9-9e50ea6e0079"], "files": ["usc_trossen/video_000009.mp4", "usc_trossen/video_000005.mp4"], "task_norm": "Open the red drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Open the red drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/07e20f469e9047db8f71f226.mp4"}, {"type": "video", "uri": "media/robometer_ood/9e35c22104c1dd795fbf8e5e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000005-vs-video_000010/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["880498e6-ad5d-4547-9f3a-b8c41f8e5376", "fab793f7-2b04-4425-8af9-9e50ea6e0079"], "files": ["usc_trossen/video_000010.mp4", "usc_trossen/video_000005.mp4"], "task_norm": "Open the red drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Open the red drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/595ac1286fff89c431d696af.mp4"}, {"type": "video", "uri": "media/robometer_ood/2f84958616ac4373bda9d3fc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000006-vs-video_000008/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["4798c411-caa3-405e-9afa-132d0a2617f1", "6eeb8f4c-2929-4b6f-a85a-2038b6302784"], "files": ["usc_trossen/video_000008.mp4", "usc_trossen/video_000006.mp4"], "task_norm": "Open the red drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Open the red drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ae7ef8156d8ea6c2984b11f1.mp4"}, {"type": "video", "uri": "media/robometer_ood/e859ecc1537c6e80f5ff4f15.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000007-vs-video_000011/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["c8613f0d-474c-46ec-958d-f3bcfb132372", "017e0e20-fd15-4005-bd8e-f31f8e2350c5"], "files": ["usc_trossen/video_000007.mp4", "usc_trossen/video_000011.mp4"], "task_norm": "Open the red drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Open the red drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/07e20f469e9047db8f71f226.mp4"}, {"type": "video", "uri": "media/robometer_ood/595ac1286fff89c431d696af.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000010-vs-video_000008/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["880498e6-ad5d-4547-9f3a-b8c41f8e5376", "4798c411-caa3-405e-9afa-132d0a2617f1"], "files": ["usc_trossen/video_000010.mp4", "usc_trossen/video_000008.mp4"], "task_norm": "Open the red drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Open the red drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8a7cb858a552adc296e98610.mp4"}, {"type": "video", "uri": "media/robometer_ood/e859ecc1537c6e80f5ff4f15.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000011-vs-video_000009/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["fcdd7b74-542e-455f-aca8-f6bd110144a8", "017e0e20-fd15-4005-bd8e-f31f8e2350c5"], "files": ["usc_trossen/video_000009.mp4", "usc_trossen/video_000011.mp4"], "task_norm": "Open the red drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 152, "role": "preference"}}
{"state": {"task": "Remove the lid from the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/32d1cdb85ab9e2250d8dc454.mp4"}, {"type": "video", "uri": "media/robometer_ood/8c451e965a0c3847606738d9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000013-vs-video_000012/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Remove the lid from the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["8d2311da-1e57-4117-b6fe-52332558902b", "63d1a6c1-c440-4562-afa6-f8dd2d8d0a52"], "files": ["usc_trossen/video_000013.mp4", "usc_trossen/video_000012.mp4"], "task_norm": "Remove the lid from the pot", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 154, "role": "preference"}}
{"state": {"task": "Remove the lid from the pot", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3262adb92cb76178a6f730bb.mp4"}, {"type": "video", "uri": "media/robometer_ood/6ec00c76168c9e2db82bc163.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000014-vs-video_000015/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Remove the lid from the pot", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["08dec425-2969-4989-9c40-8468babfad1d", "982f175d-6da5-4070-a52e-53681f28a5f2"], "files": ["usc_trossen/video_000014.mp4", "usc_trossen/video_000015.mp4"], "task_norm": "Remove the lid from the pot", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 154, "role": "preference"}}
{"state": {"task": "Unzip the pencil case", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5c8ae78c938b81ece001df70.mp4"}, {"type": "video", "uri": "media/robometer_ood/72d1ef3a1d61c4171950e017.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000016-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["bf3d54f4-482e-4f3a-a20f-bbcdc8c699e7", "043753c3-f59f-494f-944f-52fdd604f7b2"], "files": ["usc_trossen/video_000019.mp4", "usc_trossen/video_000016.mp4"], "task_norm": "Unzip the pencil case", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Unzip the pencil case", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/72d1ef3a1d61c4171950e017.mp4"}, {"type": "video", "uri": "media/robometer_ood/86ecee5ad2e5a67cf3029f2f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000016-vs-video_000023/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["043753c3-f59f-494f-944f-52fdd604f7b2", "9af656e8-3629-4c28-9858-89a6105488e6"], "files": ["usc_trossen/video_000016.mp4", "usc_trossen/video_000023.mp4"], "task_norm": "Unzip the pencil case", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Unzip the pencil case", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/762302281e04f3e3ddf8bec5.mp4"}, {"type": "video", "uri": "media/robometer_ood/86ecee5ad2e5a67cf3029f2f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000023-vs-video_000022/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["1372e792-8b26-41a5-aa02-01528fc941a2", "9af656e8-3629-4c28-9858-89a6105488e6"], "files": ["usc_trossen/video_000022.mp4", "usc_trossen/video_000023.mp4"], "task_norm": "Unzip the pencil case", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Unzip the pencil case", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/111ef9aec7f3a4c7bb582ca0.mp4"}, {"type": "video", "uri": "media/robometer_ood/762302281e04f3e3ddf8bec5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000024-vs-video_000022/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["e8b107bc-985d-4093-85a9-fa3e4a88eeaa", "1372e792-8b26-41a5-aa02-01528fc941a2"], "files": ["usc_trossen/video_000024.mp4", "usc_trossen/video_000022.mp4"], "task_norm": "Unzip the pencil case", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Unzip the pencil case", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5c8ae78c938b81ece001df70.mp4"}, {"type": "video", "uri": "media/robometer_ood/84e45f5c4429e66fb1a1b451.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000025-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["bf3d54f4-482e-4f3a-a20f-bbcdc8c699e7", "07948cb2-2c2f-470e-8c28-cb68eb462655"], "files": ["usc_trossen/video_000019.mp4", "usc_trossen/video_000025.mp4"], "task_norm": "Unzip the pencil case", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Unzip the pencil case", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9e676400acb0c58b82582335.mp4"}, {"type": "video", "uri": "media/robometer_ood/84e45f5c4429e66fb1a1b451.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000026-vs-video_000025/robometer_traj_pref", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["cd11463a-39eb-45ce-9bb9-0b9f76e4f63d", "07948cb2-2c2f-470e-8c28-cb68eb462655"], "files": ["usc_trossen/video_000026.mp4", "usc_trossen/video_000025.mp4"], "task_norm": "Unzip the pencil case", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Close the drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/159c61767381c90e054cdd51.mp4"}, {"type": "video", "uri": "media/robometer_ood/7251ca8acc07088ceab3e58e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000002-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["9f99930c-cbf8-46b0-882f-ff7b3fcc4e93", "2a2fb66b-b10d-44df-9869-bb60cfd76420"], "files": ["usc_xarm/video_000000.mp4", "usc_xarm/video_000002.mp4"], "task_norm": "Close the drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Close the drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bf3b569b5f2c2ea6d7564c27.mp4"}, {"type": "video", "uri": "media/robometer_ood/ac6e7259397b2dda9bfefd1c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000003-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["4a290bd7-6348-4ceb-af1e-3caada5c63ee", "8d1d7235-b515-47b8-b8e8-c8d913d85ab7"], "files": ["usc_xarm/video_000003.mp4", "usc_xarm/video_000001.mp4"], "task_norm": "Close the drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Close the drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/159c61767381c90e054cdd51.mp4"}, {"type": "video", "uri": "media/robometer_ood/e7c46f757a9f94613489f630.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000004-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["9f99930c-cbf8-46b0-882f-ff7b3fcc4e93", "4f6262d5-427f-4fb1-9e88-0bf7c319e434"], "files": ["usc_xarm/video_000000.mp4", "usc_xarm/video_000004.mp4"], "task_norm": "Close the drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Close the drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e7c46f757a9f94613489f630.mp4"}, {"type": "video", "uri": "media/robometer_ood/7251ca8acc07088ceab3e58e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000004-vs-video_000002/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["4f6262d5-427f-4fb1-9e88-0bf7c319e434", "2a2fb66b-b10d-44df-9869-bb60cfd76420"], "files": ["usc_xarm/video_000004.mp4", "usc_xarm/video_000002.mp4"], "task_norm": "Close the drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Close the drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/aebe876dceec2ba599c6e1c6.mp4"}, {"type": "video", "uri": "media/robometer_ood/ac6e7259397b2dda9bfefd1c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000005-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["8ddf2911-58ad-46ba-87cd-7710f2920fbf", "8d1d7235-b515-47b8-b8e8-c8d913d85ab7"], "files": ["usc_xarm/video_000005.mp4", "usc_xarm/video_000001.mp4"], "task_norm": "Close the drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Close the drawer", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/aebe876dceec2ba599c6e1c6.mp4"}, {"type": "video", "uri": "media/robometer_ood/bf3b569b5f2c2ea6d7564c27.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000005-vs-video_000003/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["8ddf2911-58ad-46ba-87cd-7710f2920fbf", "4a290bd7-6348-4ceb-af1e-3caada5c63ee"], "files": ["usc_xarm/video_000005.mp4", "usc_xarm/video_000003.mp4"], "task_norm": "Close the drawer", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/caf81c060073a3b6dc1e170a.mp4"}, {"type": "video", "uri": "media/robometer_ood/a5c65de34c0aac6903353ba4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000008-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["ea8f302b-f3d2-4a07-8d5d-618028b6fe96", "73b0fb7d-dc98-4a97-a39e-fe99acac7b2c"], "files": ["usc_xarm/video_000008.mp4", "usc_xarm/video_000007.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c3729cd28a5c60b40a180ac6.mp4"}, {"type": "video", "uri": "media/robometer_ood/c791d5e90a895046d4c48ad3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000009-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["792655c7-6a7f-42a5-b058-70f88917e3b3", "887bb1d0-ec5a-4f9d-844f-56f95a787790"], "files": ["usc_xarm/video_000009.mp4", "usc_xarm/video_000006.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a5c65de34c0aac6903353ba4.mp4"}, {"type": "video", "uri": "media/robometer_ood/e60d984ae0bbb741237ae44b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000010-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["73b0fb7d-dc98-4a97-a39e-fe99acac7b2c", "e5b8e128-8655-4f86-b68c-0e5d8647c68e"], "files": ["usc_xarm/video_000007.mp4", "usc_xarm/video_000010.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/caf81c060073a3b6dc1e170a.mp4"}, {"type": "video", "uri": "media/robometer_ood/e60d984ae0bbb741237ae44b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000010-vs-video_000008/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["ea8f302b-f3d2-4a07-8d5d-618028b6fe96", "e5b8e128-8655-4f86-b68c-0e5d8647c68e"], "files": ["usc_xarm/video_000008.mp4", "usc_xarm/video_000010.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c791d5e90a895046d4c48ad3.mp4"}, {"type": "video", "uri": "media/robometer_ood/6de0729023c25daffa3f9577.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000011-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["887bb1d0-ec5a-4f9d-844f-56f95a787790", "71847ff7-3199-4794-bc44-373cc5315777"], "files": ["usc_xarm/video_000006.mp4", "usc_xarm/video_000011.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6de0729023c25daffa3f9577.mp4"}, {"type": "video", "uri": "media/robometer_ood/c3729cd28a5c60b40a180ac6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000011-vs-video_000009/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["71847ff7-3199-4794-bc44-373cc5315777", "792655c7-6a7f-42a5-b058-70f88917e3b3"], "files": ["usc_xarm/video_000011.mp4", "usc_xarm/video_000009.mp4"], "task_norm": "Fold the towel in half", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Put blue cup in sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fad73ee6d8815d2d11b8e41f.mp4"}, {"type": "video", "uri": "media/robometer_ood/af395c937ec4c0e650a5d00a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000014-vs-video_000012/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["0bd8cd2f-85b2-4143-992a-2e185c502103", "897bdcd4-24ff-4795-806a-c7ca3808e312"], "files": ["usc_xarm/video_000012.mp4", "usc_xarm/video_000014.mp4"], "task_norm": "Put blue cup in sink", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Put blue cup in sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5b3a2516c808ba67a8f6c7ff.mp4"}, {"type": "video", "uri": "media/robometer_ood/39d40122ad1c313a09b245a2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000015-vs-video_000013/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["dab34346-c997-4a7b-b4c3-28f8124d0427", "3750bd8b-76b1-4f2e-a862-ac9f6272a2c7"], "files": ["usc_xarm/video_000013.mp4", "usc_xarm/video_000015.mp4"], "task_norm": "Put blue cup in sink", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Put blue cup in sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5607241f660c236fa1c4ab1f.mp4"}, {"type": "video", "uri": "media/robometer_ood/fad73ee6d8815d2d11b8e41f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000016-vs-video_000012/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["84b5e7d9-7f36-4148-a635-196c90eb2b26", "0bd8cd2f-85b2-4143-992a-2e185c502103"], "files": ["usc_xarm/video_000016.mp4", "usc_xarm/video_000012.mp4"], "task_norm": "Put blue cup in sink", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Put blue cup in sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/af395c937ec4c0e650a5d00a.mp4"}, {"type": "video", "uri": "media/robometer_ood/5607241f660c236fa1c4ab1f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000016-vs-video_000014/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["897bdcd4-24ff-4795-806a-c7ca3808e312", "84b5e7d9-7f36-4148-a635-196c90eb2b26"], "files": ["usc_xarm/video_000014.mp4", "usc_xarm/video_000016.mp4"], "task_norm": "Put blue cup in sink", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Put blue cup in sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5b3a2516c808ba67a8f6c7ff.mp4"}, {"type": "video", "uri": "media/robometer_ood/f0c3fb6e3f5b02d691dce852.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000017-vs-video_000013/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["dab34346-c997-4a7b-b4c3-28f8124d0427", "bbd59642-b65c-4f8d-9ab6-5ae9f5fe73cd"], "files": ["usc_xarm/video_000013.mp4", "usc_xarm/video_000017.mp4"], "task_norm": "Put blue cup in sink", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Put blue cup in sink", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/39d40122ad1c313a09b245a2.mp4"}, {"type": "video", "uri": "media/robometer_ood/f0c3fb6e3f5b02d691dce852.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000017-vs-video_000015/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["3750bd8b-76b1-4f2e-a862-ac9f6272a2c7", "bbd59642-b65c-4f8d-9ab6-5ae9f5fe73cd"], "files": ["usc_xarm/video_000015.mp4", "usc_xarm/video_000017.mp4"], "task_norm": "Put blue cup in sink", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 153, "role": "preference"}}
{"state": {"task": "Put the banana on the red plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7685bfe0990f9426409242c7.mp4"}, {"type": "video", "uri": "media/robometer_ood/ab402a50c949b045820a1c11.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000020-vs-video_000018/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["f7dea129-57fa-4f5d-8134-679626274759", "1f39624c-ffef-436e-b0df-0b4b4de6ebff"], "files": ["usc_xarm/video_000018.mp4", "usc_xarm/video_000020.mp4"], "task_norm": "Put the banana on the red plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the banana on the red plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5b5d2f772189aa4f816dfae7.mp4"}, {"type": "video", "uri": "media/robometer_ood/aca13fb1fdad48ca550dd3ad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000021-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["5244aaaa-04a0-4ccd-bbad-29b2e6c8023c", "8ddd569c-f15f-4d22-abbc-42f877856a7a"], "files": ["usc_xarm/video_000019.mp4", "usc_xarm/video_000021.mp4"], "task_norm": "Put the banana on the red plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the banana on the red plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/54358380194193aa8056daf3.mp4"}, {"type": "video", "uri": "media/robometer_ood/7685bfe0990f9426409242c7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000022-vs-video_000018/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["fb7d331c-0c73-4856-836f-42576128d95b", "f7dea129-57fa-4f5d-8134-679626274759"], "files": ["usc_xarm/video_000022.mp4", "usc_xarm/video_000018.mp4"], "task_norm": "Put the banana on the red plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the banana on the red plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/54358380194193aa8056daf3.mp4"}, {"type": "video", "uri": "media/robometer_ood/ab402a50c949b045820a1c11.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000022-vs-video_000020/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["fb7d331c-0c73-4856-836f-42576128d95b", "1f39624c-ffef-436e-b0df-0b4b4de6ebff"], "files": ["usc_xarm/video_000022.mp4", "usc_xarm/video_000020.mp4"], "task_norm": "Put the banana on the red plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the banana on the red plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9db2d885d74bd1599a325706.mp4"}, {"type": "video", "uri": "media/robometer_ood/5b5d2f772189aa4f816dfae7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000023-vs-video_000019/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["ffc600a7-d58a-40e9-b4c4-52da5596fbf8", "5244aaaa-04a0-4ccd-bbad-29b2e6c8023c"], "files": ["usc_xarm/video_000023.mp4", "usc_xarm/video_000019.mp4"], "task_norm": "Put the banana on the red plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the banana on the red plate", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9db2d885d74bd1599a325706.mp4"}, {"type": "video", "uri": "media/robometer_ood/aca13fb1fdad48ca550dd3ad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000023-vs-video_000021/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["ffc600a7-d58a-40e9-b4c4-52da5596fbf8", "8ddd569c-f15f-4d22-abbc-42f877856a7a"], "files": ["usc_xarm/video_000023.mp4", "usc_xarm/video_000021.mp4"], "task_norm": "Put the banana on the red plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the eggplant in the black bowl", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6fba71d11993ea2c64f81d12.mp4"}, {"type": "video", "uri": "media/robometer_ood/15070d458035079342c0c0bc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000026-vs-video_000025/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["0947d5d3-1346-472e-9640-4e5c4f9141b7", "717ad51f-4bfb-4657-bf2a-01fbb3f072eb"], "files": ["usc_xarm/video_000025.mp4", "usc_xarm/video_000026.mp4"], "task_norm": "Put the eggplant in the black bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the eggplant in the black bowl", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7c95def3806e1384481c126d.mp4"}, {"type": "video", "uri": "media/robometer_ood/eada55b4dc588560761fedf2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000027-vs-video_000024/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["0bef6574-3f66-4f87-96f6-24a29e470195", "ed87a481-15e0-4cbb-bf2c-05e8bd070c73"], "files": ["usc_xarm/video_000027.mp4", "usc_xarm/video_000024.mp4"], "task_norm": "Put the eggplant in the black bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the eggplant in the black bowl", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/15d1796342ba4f840af450db.mp4"}, {"type": "video", "uri": "media/robometer_ood/eada55b4dc588560761fedf2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000028-vs-video_000024/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["312f4b8a-6633-43bd-9094-e6ec964d609a", "ed87a481-15e0-4cbb-bf2c-05e8bd070c73"], "files": ["usc_xarm/video_000028.mp4", "usc_xarm/video_000024.mp4"], "task_norm": "Put the eggplant in the black bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the eggplant in the black bowl", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/15d1796342ba4f840af450db.mp4"}, {"type": "video", "uri": "media/robometer_ood/15070d458035079342c0c0bc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000028-vs-video_000026/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["312f4b8a-6633-43bd-9094-e6ec964d609a", "717ad51f-4bfb-4657-bf2a-01fbb3f072eb"], "files": ["usc_xarm/video_000028.mp4", "usc_xarm/video_000026.mp4"], "task_norm": "Put the eggplant in the black bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the eggplant in the black bowl", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5ca1d638d6b2c79f4be8e684.mp4"}, {"type": "video", "uri": "media/robometer_ood/6fba71d11993ea2c64f81d12.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000029-vs-video_000025/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["70cb5268-c3e6-4e87-99a1-95aa4ffe2899", "0947d5d3-1346-472e-9640-4e5c4f9141b7"], "files": ["usc_xarm/video_000029.mp4", "usc_xarm/video_000025.mp4"], "task_norm": "Put the eggplant in the black bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the eggplant in the black bowl", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7c95def3806e1384481c126d.mp4"}, {"type": "video", "uri": "media/robometer_ood/5ca1d638d6b2c79f4be8e684.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000029-vs-video_000027/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["0bef6574-3f66-4f87-96f6-24a29e470195", "70cb5268-c3e6-4e87-99a1-95aa4ffe2899"], "files": ["usc_xarm/video_000027.mp4", "usc_xarm/video_000029.mp4"], "task_norm": "Put the eggplant in the black bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the green shape on the red cylinder", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/13841d846ef9da3a364a595e.mp4"}, {"type": "video", "uri": "media/robometer_ood/9087f69ade3486e409433748.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000032-vs-video_000031/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["f61a687b-c4d2-4bb8-9bc6-22d922c891e7", "1454909a-f782-4dbc-909c-3b4680d4ccd8"], "files": ["usc_xarm/video_000031.mp4", "usc_xarm/video_000032.mp4"], "task_norm": "Stack the green shape on the red cylinder", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the green shape on the red cylinder", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/41dba0e9c4c8d75b0fde3500.mp4"}, {"type": "video", "uri": "media/robometer_ood/aec2077a844db982eed565e7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000033-vs-video_000030/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["64e17159-9d91-4147-bfe6-fe19c037f1b6", "83684337-cf58-4b1b-89cc-b3073d268d92"], "files": ["usc_xarm/video_000030.mp4", "usc_xarm/video_000033.mp4"], "task_norm": "Stack the green shape on the red cylinder", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the green shape on the red cylinder", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5bb05087790941c856429561.mp4"}, {"type": "video", "uri": "media/robometer_ood/13841d846ef9da3a364a595e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000034-vs-video_000031/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["96fbffd6-97ce-464c-a63e-75f96c745592", "f61a687b-c4d2-4bb8-9bc6-22d922c891e7"], "files": ["usc_xarm/video_000034.mp4", "usc_xarm/video_000031.mp4"], "task_norm": "Stack the green shape on the red cylinder", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the green shape on the red cylinder", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9087f69ade3486e409433748.mp4"}, {"type": "video", "uri": "media/robometer_ood/5bb05087790941c856429561.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000034-vs-video_000032/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["1454909a-f782-4dbc-909c-3b4680d4ccd8", "96fbffd6-97ce-464c-a63e-75f96c745592"], "files": ["usc_xarm/video_000032.mp4", "usc_xarm/video_000034.mp4"], "task_norm": "Stack the green shape on the red cylinder", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the green shape on the red cylinder", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c8ff7bfc1f5425bb0e9a9db6.mp4"}, {"type": "video", "uri": "media/robometer_ood/41dba0e9c4c8d75b0fde3500.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000035-vs-video_000030/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["ae38a588-1a6e-433b-8317-fff8fbcee422", "64e17159-9d91-4147-bfe6-fe19c037f1b6"], "files": ["usc_xarm/video_000035.mp4", "usc_xarm/video_000030.mp4"], "task_norm": "Stack the green shape on the red cylinder", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the green shape on the red cylinder", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c8ff7bfc1f5425bb0e9a9db6.mp4"}, {"type": "video", "uri": "media/robometer_ood/aec2077a844db982eed565e7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000035-vs-video_000033/robometer_traj_pref", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["ae38a588-1a6e-433b-8317-fff8fbcee422", "83684337-cf58-4b1b-89cc-b3073d268d92"], "files": ["usc_xarm/video_000035.mp4", "usc_xarm/video_000033.mp4"], "task_norm": "Stack the green shape on the red cylinder", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the bread in the oven.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e2fad2460c1dad7372b2de59.mp4"}, {"type": "video", "uri": "media/robometer_ood/72e6dced54a807dd8463b069.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000008-vs-video_000003/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the bread in the oven", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["83cbb89e-6649-4456-9d93-9a50670e51b3", "b663bd06-44f9-4fb2-a411-de51b99c2fc6"], "files": ["utd_so101/video_000003.mp4", "utd_so101/video_000008.mp4"], "task_norm": "Put the bread in the oven", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 154, "role": "preference"}}
{"state": {"task": "Press the button.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4dd0004ad60a52433a42071b.mp4"}, {"type": "video", "uri": "media/robometer_ood/e47ab17188ac86bf6e8948b2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000009-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Press the button", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["a65c9878-0537-4750-954b-b424b50bf24c", "c8ec8598-f543-4765-9b0a-f573d0fd0b88"], "files": ["utd_so101/video_000009.mp4", "utd_so101/video_000000.mp4"], "task_norm": "Press the button", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Put the marker into the pen cup.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e7288c0900b91ceab3a27381.mp4"}, {"type": "video", "uri": "media/robometer_ood/3dc02fa38305ae439fdfe778.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000012-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the marker into the pen cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["72c73b9d-bc33-4181-804d-bb3096034da4", "cb283639-f8f0-436b-a1ee-94e49bd69d7b"], "files": ["utd_so101/video_000001.mp4", "utd_so101/video_000012.mp4"], "task_norm": "Put the marker into the pen cup", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the red bowl on the blue plate.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/667ab634edd7f64eb044dbba.mp4"}, {"type": "video", "uri": "media/robometer_ood/8cb57cb953ce1ed36bc7495a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000013-vs-video_000002/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the red bowl on the blue plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["a5bdb049-3422-4fbb-bfe7-2c637057a6fd", "6764ffec-e78a-41de-8f14-68b4d8ae8fbc"], "files": ["utd_so101/video_000002.mp4", "utd_so101/video_000013.mp4"], "task_norm": "Put the red bowl on the blue plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Collect the fork to the yellow box.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e6d4fac90e6ed7065aa22172.mp4"}, {"type": "video", "uri": "media/robometer_ood/0abd353607ecf5bdc45effff.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000014-vs-video_000004/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Collect the fork to the yellow box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["b75e38bb-49db-4655-bd85-09802cf5c7cb", "77f40402-ef04-45b1-ae7c-9be90a6d80dc"], "files": ["utd_so101/video_000014.mp4", "utd_so101/video_000004.mp4"], "task_norm": "Collect the fork to the yellow box", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the pear in the yellow plate.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3386120d9e28bfc3fb7ae451.mp4"}, {"type": "video", "uri": "media/robometer_ood/f242921a67eefbd7d5967ac3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000015-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the pear in the yellow plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["d159b432-2747-4618-90cd-bf4bfbf8145e", "65d1db25-46ce-4175-8b52-fffb517b49bf"], "files": ["utd_so101/video_000015.mp4", "utd_so101/video_000006.mp4"], "task_norm": "Put the pear in the yellow plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Stack the green block on the red block.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/50ed89f40ba07519345ca253.mp4"}, {"type": "video", "uri": "media/robometer_ood/e02d83f69d44eac663f3a41e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000016-vs-video_000005/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Stack the green block on the red block", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["d16d47de-7b92-4d6e-8e4c-895bc8357a65", "805de0a8-f4cf-490d-8e2f-bc6cb70a0130"], "files": ["utd_so101/video_000016.mp4", "utd_so101/video_000005.mp4"], "task_norm": "Stack the green block on the red block", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Add pepper to the green bowl.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7fe2550e3a71a4bd8874e107.mp4"}, {"type": "video", "uri": "media/robometer_ood/1315ec7b9cedc6721e9e43e2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000017-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Add pepper to the green bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["049852cd-8733-4132-b486-5390b646768a", "af70eeb2-e86d-45b7-a5fb-a2793c736072"], "files": ["utd_so101/video_000017.mp4", "utd_so101/video_000007.mp4"], "task_norm": "Add pepper to the green bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 154, "role": "preference"}}
{"state": {"task": "Put the eraser to the blue pencil box.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0c5cd5fdf76a249b1ece9a0a.mp4"}, {"type": "video", "uri": "media/robometer_ood/5b3b1b6bdfcf39cf403edc50.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000018-vs-video_000010/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the eraser to the blue pencil box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "failure", "quality_B": "suboptimal", "native_ids": ["e838f14d-2c55-45e2-9e00-d8883880e9af", "b6a91500-67d1-4bb2-8a5b-ca2cd8dd7596"], "files": ["utd_so101/video_000010.mp4", "utd_so101/video_000018.mp4"], "task_norm": "Put the eraser to the blue pencil box", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 157, "role": "preference"}}
{"state": {"task": "Press the button.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ee9f94f244bca1673fb8afd0.mp4"}, {"type": "video", "uri": "media/robometer_ood/e47ab17188ac86bf6e8948b2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000019-vs-video_000000/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Press the button", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["5296a584-bd36-43a3-9b64-58980b7814da", "c8ec8598-f543-4765-9b0a-f573d0fd0b88"], "files": ["utd_so101/video_000019.mp4", "utd_so101/video_000000.mp4"], "task_norm": "Press the button", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Press the button.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ee9f94f244bca1673fb8afd0.mp4"}, {"type": "video", "uri": "media/robometer_ood/4dd0004ad60a52433a42071b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000019-vs-video_000009/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Press the button", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["5296a584-bd36-43a3-9b64-58980b7814da", "a65c9878-0537-4750-954b-b424b50bf24c"], "files": ["utd_so101/video_000019.mp4", "utd_so101/video_000009.mp4"], "task_norm": "Press the button", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 151, "role": "preference"}}
{"state": {"task": "Put the red cup on the purple coaster.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3b9d2d3d9737b555e0d73524.mp4"}, {"type": "video", "uri": "media/robometer_ood/7a10019f18babcbd72393737.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000020-vs-video_000011/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the red cup on the purple coaster", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "suboptimal>failure", "quality_A": "suboptimal", "quality_B": "failure", "native_ids": ["dd80bff0-4e0f-4c12-b41d-f203276e28ab", "d5c4164a-06b4-4bf0-871e-bbf6c736190a"], "files": ["utd_so101/video_000020.mp4", "utd_so101/video_000011.mp4"], "task_norm": "Put the red cup on the purple coaster", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the bread in the oven.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e2fad2460c1dad7372b2de59.mp4"}, {"type": "video", "uri": "media/robometer_ood/b594e8612baa803f1be647ad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000021-vs-video_000003/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the bread in the oven", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["83cbb89e-6649-4456-9d93-9a50670e51b3", "0641becc-c8bb-4c92-a56e-8dbd946f465e"], "files": ["utd_so101/video_000003.mp4", "utd_so101/video_000021.mp4"], "task_norm": "Put the bread in the oven", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 154, "role": "preference"}}
{"state": {"task": "Put the bread in the oven.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b594e8612baa803f1be647ad.mp4"}, {"type": "video", "uri": "media/robometer_ood/72e6dced54a807dd8463b069.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000021-vs-video_000008/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the bread in the oven", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["0641becc-c8bb-4c92-a56e-8dbd946f465e", "b663bd06-44f9-4fb2-a411-de51b99c2fc6"], "files": ["utd_so101/video_000021.mp4", "utd_so101/video_000008.mp4"], "task_norm": "Put the bread in the oven", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 154, "role": "preference"}}
{"state": {"task": "Put the red bowl on the blue plate.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/667ab634edd7f64eb044dbba.mp4"}, {"type": "video", "uri": "media/robometer_ood/c3f88849f49cf7b8c778aefe.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000022-vs-video_000002/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the red bowl on the blue plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["a5bdb049-3422-4fbb-bfe7-2c637057a6fd", "e5088127-7e1a-4526-9927-54e2e8f4b5b5"], "files": ["utd_so101/video_000002.mp4", "utd_so101/video_000022.mp4"], "task_norm": "Put the red bowl on the blue plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the red bowl on the blue plate.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c3f88849f49cf7b8c778aefe.mp4"}, {"type": "video", "uri": "media/robometer_ood/8cb57cb953ce1ed36bc7495a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000022-vs-video_000013/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the red bowl on the blue plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["e5088127-7e1a-4526-9927-54e2e8f4b5b5", "6764ffec-e78a-41de-8f14-68b4d8ae8fbc"], "files": ["utd_so101/video_000022.mp4", "utd_so101/video_000013.mp4"], "task_norm": "Put the red bowl on the blue plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the marker into the pen cup.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/008a8744f4eac6e3817a4ed2.mp4"}, {"type": "video", "uri": "media/robometer_ood/e7288c0900b91ceab3a27381.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000023-vs-video_000001/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the marker into the pen cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["dfb9e90c-75a0-4eb2-83c3-4e38cb101e0e", "72c73b9d-bc33-4181-804d-bb3096034da4"], "files": ["utd_so101/video_000023.mp4", "utd_so101/video_000001.mp4"], "task_norm": "Put the marker into the pen cup", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the marker into the pen cup.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/008a8744f4eac6e3817a4ed2.mp4"}, {"type": "video", "uri": "media/robometer_ood/3dc02fa38305ae439fdfe778.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000023-vs-video_000012/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the marker into the pen cup", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["dfb9e90c-75a0-4eb2-83c3-4e38cb101e0e", "cb283639-f8f0-436b-a1ee-94e49bd69d7b"], "files": ["utd_so101/video_000023.mp4", "utd_so101/video_000012.mp4"], "task_norm": "Put the marker into the pen cup", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the pear in the yellow plate.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f242921a67eefbd7d5967ac3.mp4"}, {"type": "video", "uri": "media/robometer_ood/1506017b3dfc4d19cf045b17.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000024-vs-video_000006/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the pear in the yellow plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["65d1db25-46ce-4175-8b52-fffb517b49bf", "2a671702-e614-4354-8098-29bc7a6b0068"], "files": ["utd_so101/video_000006.mp4", "utd_so101/video_000024.mp4"], "task_norm": "Put the pear in the yellow plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the pear in the yellow plate.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3386120d9e28bfc3fb7ae451.mp4"}, {"type": "video", "uri": "media/robometer_ood/1506017b3dfc4d19cf045b17.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000024-vs-video_000015/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the pear in the yellow plate", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["d159b432-2747-4618-90cd-bf4bfbf8145e", "2a671702-e614-4354-8098-29bc7a6b0068"], "files": ["utd_so101/video_000015.mp4", "utd_so101/video_000024.mp4"], "task_norm": "Put the pear in the yellow plate", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Stack the green block on the red block.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e02d83f69d44eac663f3a41e.mp4"}, {"type": "video", "uri": "media/robometer_ood/453f1c43077775bfb19c2640.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000025-vs-video_000005/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Stack the green block on the red block", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["805de0a8-f4cf-490d-8e2f-bc6cb70a0130", "b48899a2-969c-4fa0-9ecb-827456ff935e"], "files": ["utd_so101/video_000005.mp4", "utd_so101/video_000025.mp4"], "task_norm": "Stack the green block on the red block", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Stack the green block on the red block.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/453f1c43077775bfb19c2640.mp4"}, {"type": "video", "uri": "media/robometer_ood/50ed89f40ba07519345ca253.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000025-vs-video_000016/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Stack the green block on the red block", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["b48899a2-969c-4fa0-9ecb-827456ff935e", "d16d47de-7b92-4d6e-8e4c-895bc8357a65"], "files": ["utd_so101/video_000025.mp4", "utd_so101/video_000016.mp4"], "task_norm": "Stack the green block on the red block", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Collect the fork to the yellow box.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0abd353607ecf5bdc45effff.mp4"}, {"type": "video", "uri": "media/robometer_ood/a522022ac085bbf95256d902.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000026-vs-video_000004/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Collect the fork to the yellow box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["77f40402-ef04-45b1-ae7c-9be90a6d80dc", "17115224-7835-4a87-a7e0-f4fd6e7339f5"], "files": ["utd_so101/video_000004.mp4", "utd_so101/video_000026.mp4"], "task_norm": "Collect the fork to the yellow box", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Collect the fork to the yellow box.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a522022ac085bbf95256d902.mp4"}, {"type": "video", "uri": "media/robometer_ood/e6d4fac90e6ed7065aa22172.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000026-vs-video_000014/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Collect the fork to the yellow box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "success", "quality_B": "suboptimal", "native_ids": ["17115224-7835-4a87-a7e0-f4fd6e7339f5", "b75e38bb-49db-4655-bd85-09802cf5c7cb"], "files": ["utd_so101/video_000026.mp4", "utd_so101/video_000014.mp4"], "task_norm": "Collect the fork to the yellow box", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 155, "role": "preference"}}
{"state": {"task": "Put the eraser to the blue pencil box.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e925b99b06fff2f79db0a3c5.mp4"}, {"type": "video", "uri": "media/robometer_ood/0c5cd5fdf76a249b1ece9a0a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000027-vs-video_000010/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the eraser to the blue pencil box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["3ccee87c-e7db-41c1-9e07-b97110e95b92", "e838f14d-2c55-45e2-9e00-d8883880e9af"], "files": ["utd_so101/video_000027.mp4", "utd_so101/video_000010.mp4"], "task_norm": "Put the eraser to the blue pencil box", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 157, "role": "preference"}}
{"state": {"task": "Put the eraser to the blue pencil box.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5b3b1b6bdfcf39cf403edc50.mp4"}, {"type": "video", "uri": "media/robometer_ood/e925b99b06fff2f79db0a3c5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000027-vs-video_000018/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the eraser to the blue pencil box", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["b6a91500-67d1-4bb2-8a5b-ca2cd8dd7596", "3ccee87c-e7db-41c1-9e07-b97110e95b92"], "files": ["utd_so101/video_000018.mp4", "utd_so101/video_000027.mp4"], "task_norm": "Put the eraser to the blue pencil box", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 157, "role": "preference"}}
{"state": {"task": "Add pepper to the green bowl.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "A", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b55c0ae7816a6ca0ab34f964.mp4"}, {"type": "video", "uri": "media/robometer_ood/1315ec7b9cedc6721e9e43e2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000028-vs-video_000007/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Add pepper to the green bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "success", "quality_B": "failure", "native_ids": ["6dcbfb2a-a612-433d-a64c-c3083a32337e", "af70eeb2-e86d-45b7-a5fb-a2793c736072"], "files": ["utd_so101/video_000028.mp4", "utd_so101/video_000007.mp4"], "task_norm": "Add pepper to the green bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 154, "role": "preference"}}
{"state": {"task": "Add pepper to the green bowl.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7fe2550e3a71a4bd8874e107.mp4"}, {"type": "video", "uri": "media/robometer_ood/b55c0ae7816a6ca0ab34f964.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000028-vs-video_000017/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Add pepper to the green bowl", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["049852cd-8733-4132-b486-5390b646768a", "6dcbfb2a-a612-433d-a64c-c3083a32337e"], "files": ["utd_so101/video_000017.mp4", "utd_so101/video_000028.mp4"], "task_norm": "Add pepper to the green bowl", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 154, "role": "preference"}}
{"state": {"task": "Put the red cup on the purple coaster.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7a10019f18babcbd72393737.mp4"}, {"type": "video", "uri": "media/robometer_ood/7c65a6748fd93d21b3246f54.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000029-vs-video_000011/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the red cup on the purple coaster", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>failure", "quality_A": "failure", "quality_B": "success", "native_ids": ["d5c4164a-06b4-4bf0-871e-bbf6c736190a", "2cdf5d90-ad7d-495e-865c-bdc33dd74bf5"], "files": ["utd_so101/video_000011.mp4", "utd_so101/video_000029.mp4"], "task_norm": "Put the red cup on the purple coaster", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Put the red cup on the purple coaster.", "videos": ["rollout A", "rollout B"]}, "questions": {"decision": {"type": "choice", "instructions": "You are comparing two rollouts of a robot attempting the same manipulation task on the same setup. The state gives the task instruction; the two attached videos are rollout A and then rollout B, each from its first to its last frame. Which rollout better accomplishes the task?", "criteria": {"A": "Rollout A is better: it completes the task when B does not, or gets further toward completing it, or completes it more cleanly and directly than B.", "B": "Rollout B is better: it completes the task when A does not, or gets further toward completing it, or completes it more cleanly and directly than A."}, "label": "B", "src": "robometer_traj_pref"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3b9d2d3d9737b555e0d73524.mp4"}, {"type": "video", "uri": "media/robometer_ood/7c65a6748fd93d21b3246f54.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000029-vs-video_000020/robometer_traj_pref", "group_id": "robometer_ood/utd_so101/Put the red cup on the purple coaster", "family": "robometer_traj_pref", "pairwise": {"media_index": [0, 1]}, "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "level_pair": "success>suboptimal", "quality_A": "suboptimal", "quality_B": "success", "native_ids": ["dd80bff0-4e0f-4c12-b41d-f203276e28ab", "2cdf5d90-ad7d-495e-865c-bdc33dd74bf5"], "files": ["utd_so101/video_000020.mp4", "utd_so101/video_000029.mp4"], "task_norm": "Put the red cup on the purple coaster", "video_info": [{"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}], "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames); video 1 = rollout A, video 2 = rollout B"}, "n_media": 2, "approx_tokens": 156, "role": "preference"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/10d2670f4c5d94a15f755a69.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000024/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "9004622a-b59a-4aab-b02c-6d0ac4c4bf79", "file": "mit_franka/video_000024.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/279067a6ee71b3bd6541c6f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000025/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "000c9dd6-e04c-4ee5-8af4-3abf3314c673", "file": "mit_franka/video_000025.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1b00e8d6604d7fa73693df94.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000026/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "7a6d915b-9383-4c99-ac43-30b3e6ddb340", "file": "mit_franka/video_000026.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cbc8127ed56bc4cf45c2f8fd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000027/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "14a26a41-f949-4c2a-aced-4d5d0b9f6f6f", "file": "mit_franka/video_000027.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/22d772e51515af4e88a7df8a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000028/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "c0806df4-7289-4c26-be73-40b4a33b471b", "file": "mit_franka/video_000028.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a3b994fe7be669f93bbf4bbe.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000029/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "01e3be9a-96b0-45b4-a15d-9827a50ba098", "file": "mit_franka/video_000029.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/715f5ef324357037d32ec950.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000030/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "faf4bcb1-d384-403c-829f-53b965fcd58d", "file": "mit_franka/video_000030.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/27c4f48937bfd2b505635e0f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000031/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8c333dd9-3d99-448e-ae58-b681b36d502b", "file": "mit_franka/video_000031.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/75e35dba7d2bc3f3a640c2dd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000032/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "45e39a64-fc77-4650-844d-8a2d82dc934d", "file": "mit_franka/video_000032.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/938ae535abbbe185a71632bf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000033/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "741f8ab9-6d38-457e-95c9-4e52a1605535", "file": "mit_franka/video_000033.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c14e604a12a1b0b7980bb45f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000034/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "bf994e7d-3f4a-40c4-9da8-cd464d98f373", "file": "mit_franka/video_000034.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b2991ab6bbd57a978f8fa4dc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000035/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "6b423129-f607-4ae2-8e84-4cf98ba5b7fa", "file": "mit_franka/video_000035.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3de5c58d4f2cb4c474951184.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000036/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ec5cc6e4-1e92-4d88-8a88-9b82a40f51ce", "file": "mit_franka/video_000036.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/78676aff96bb3d6aa4ab2d78.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000037/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "05538a11-c296-466f-8865-e4e3c0ea5d58", "file": "mit_franka/video_000037.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0950432353cf90624731a736.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000038/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "07091ac9-b6b3-4a05-bf89-f414182ed8a8", "file": "mit_franka/video_000038.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/65d45d1162d7ea79bd1a2859.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000039/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "dd34d3ce-3758-4543-8bd1-c10135b46e71", "file": "mit_franka/video_000039.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b71667f656ada7742b705293.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000040/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8beeb483-c9bd-4e6b-a2a7-fba7557db07e", "file": "mit_franka/video_000040.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b4aacbdd9888a547c8c8b39a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000041/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "49163975-14fa-4df9-b022-361889a59960", "file": "mit_franka/video_000041.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4cc8b569b9a1f65971e6825f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000042/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "7603a1bc-a20a-412c-92fa-76e4958ad58f", "file": "mit_franka/video_000042.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b259c7110007e6b27d122d42.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000043/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "519d5f5d-7b81-4b7f-9134-d14af6922ab3", "file": "mit_franka/video_000043.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2777ee313d5bd4cbd9e93735.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000044/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ef82095d-6647-438b-ae41-33b46b745009", "file": "mit_franka/video_000044.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5bc5f0483e6a93e5b5709cfe.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000045/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "815000fd-8761-4399-a7e3-c57e20f82bc5", "file": "mit_franka/video_000045.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bcf2d84744fa69263e33af9a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000046/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "c77fedd0-526f-40a8-9a00-fd24fb4f450c", "file": "mit_franka/video_000046.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/17cb21b7b907e84744bdcc2f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000047/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b8821982-69c0-4dcd-bdeb-ddce09d47295", "file": "mit_franka/video_000047.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5b88762414ab07b7d55f5e1f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000048/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "9f5b84f6-26f9-40b1-8bb4-a133beb51837", "file": "mit_franka/video_000048.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ad135ee3cfa289ef1f6a8b67.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000049/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "914335e2-47c0-4c8c-9a27-59cff97f9807", "file": "mit_franka/video_000049.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4e785d76f0c123c39c4d4d13.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000050/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "e55a6b32-2e81-47fd-bf93-7d455f587f8d", "file": "mit_franka/video_000050.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fc6f74b59c608850b1930e37.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000051/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8d12f346-fc65-40a8-8ffe-131ef8a43d2c", "file": "mit_franka/video_000051.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1fe5f3eeb83147302274e686.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000052/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "c7cfb350-f106-49f6-a3b5-ab0799faaf08", "file": "mit_franka/video_000052.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8e25ef44cda2d5c69c6124e9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000053/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d12139fb-0bea-46b1-abfd-0dc23e000fa5", "file": "mit_franka/video_000053.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0e052d4e5ceab0c77630395e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000054/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "6b5f3b7d-4869-455a-a689-432dceeca830", "file": "mit_franka/video_000054.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f7df15d61332db89c351712a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000057/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "12ae16fd-d749-4a57-b6c3-9b8d4f08749d", "file": "mit_franka/video_000057.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c1e08aa7ceb16cbe244e48bd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000086/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "20b51c66-336b-42df-a619-da72b8501547", "file": "mit_franka/video_000086.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6fd1eb026a7026e95190cbf9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000087/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b0a316a8-9cad-428f-a3e6-0662ff0412dd", "file": "mit_franka/video_000087.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b9f3708f551b5030e763f909.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000089/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "07bd8aa3-9565-4931-a1ae-2b57622a89eb", "file": "mit_franka/video_000089.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f667ff391a6dc25fd532fac5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000091/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "eb51e61f-1866-4fb9-814a-12b2f5a44298", "file": "mit_franka/video_000091.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/51df25f0dc82b86f12ea658a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000092/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "f5b509e0-b261-4452-83e4-62683008d0f4", "file": "mit_franka/video_000092.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1daa40e150a06f3fe9fa1498.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000093/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2ebe5d8d-a81f-4626-95de-1941c78f3a23", "file": "mit_franka/video_000093.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/33b18a16f1791d24c2859ddd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000094/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "6a963be3-befc-454d-907a-d6634557e682", "file": "mit_franka/video_000094.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0c4df98d982334e81b275a46.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000095/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "01baeffb-0ef4-4921-994f-e1c52a902d09", "file": "mit_franka/video_000095.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fa50f8ca47ce609f0ae567f9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000096/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "3677a3d6-b3b4-485e-a126-f6bb39b2ddbc", "file": "mit_franka/video_000096.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0c6ceecef563b31cf2939f91.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000097/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2f3fb3c1-4d80-4737-8da3-6113c4566ab6", "file": "mit_franka/video_000097.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/85efd00b00b26190ea099729.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000098/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8711a72c-b077-487d-82f5-eb271a00e839", "file": "mit_franka/video_000098.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3ddf31acacfc6e6da329cfa9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000099/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b5dd5264-67f9-41c0-8aa6-07a103b19e23", "file": "mit_franka/video_000099.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2633672ecc4feb34e1c68628.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000100/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "83ccff35-571c-4a66-bf31-9d42097f5e1a", "file": "mit_franka/video_000100.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/32f9ebfc6076e809ba0809f3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000101/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "7e4c9f98-39c1-4984-b86b-4f60d9eec5a5", "file": "mit_franka/video_000101.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e02755a09d50f2a169fe7ca0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000102/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "915d82b0-fb49-4257-8c60-668ead0af980", "file": "mit_franka/video_000102.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1876782203f1efab0706abcf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000103/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "e25bd65b-1441-4803-8bd0-266526d4396b", "file": "mit_franka/video_000103.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8fccc30dba7b991c22412112.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000104/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "42965aef-8c98-4cc9-a49c-d7f8a3a6d86e", "file": "mit_franka/video_000104.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f402cebaac8d6c8e69893b47.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000105/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "52b30f87-5541-4d97-976c-f1083c06764d", "file": "mit_franka/video_000105.mp4", "quality_label": "failure", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7af101e6cae6d9b79078761a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000106/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "1de13c1d-4b72-437f-a1ed-4050f85f6557", "file": "mit_franka/video_000106.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/af9785883c3dfc131a48ab10.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000107/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "eeb8b278-c25d-43bb-9df5-1813d46886cc", "file": "mit_franka/video_000107.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b2972d784d8515def4304a96.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000108/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "1975e479-961e-459e-be6f-5eae24900626", "file": "mit_franka/video_000108.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/af0ee5e7fd1b9327409cc8f8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000109/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "4596a95d-7900-4b36-b7ba-5782fab9b8c9", "file": "mit_franka/video_000109.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d6501b513d0e53d18090c5a6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000110/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "6a94664c-7aa5-4936-9c88-40be9319a1b6", "file": "mit_franka/video_000110.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0ab9289f7eab40cf0255bab0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000111/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "35fa519b-3bbd-40ff-8b35-3d860eee6df2", "file": "mit_franka/video_000111.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b233aa87e62f212d6aae0763.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000112/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "0ee0335f-aa12-44dd-acd1-8344c231604b", "file": "mit_franka/video_000112.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7486f21678e7d69d1d055835.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000113/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d92c5b5a-f552-4d0b-be41-091a5525149f", "file": "mit_franka/video_000113.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/946b8bb8c3d1403c723b9c78.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000114/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "30bffa61-10ef-4870-bc86-24ec5e05f2ed", "file": "mit_franka/video_000114.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a1b770556be87e25ee17aea9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000115/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "538df936-e738-4213-a454-c48cfd0ddff5", "file": "mit_franka/video_000115.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d3d02ad68a051f6d9595291d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000116/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "4a81a138-b2ea-43e1-8a1c-1fc7497dbe3f", "file": "mit_franka/video_000116.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b3339b395f0d59f3d7d0adbe.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000117/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "f1e72daf-47b9-4017-9de1-8b760e30ba75", "file": "mit_franka/video_000117.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/baf65f48af730f6e7c91da96.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000118/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "cf62479b-dec8-4b56-903d-9c66cf82e972", "file": "mit_franka/video_000118.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a0f3beed83982d4cfcddfa7b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000119/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ede2ccec-63de-4297-9e03-c51508b83211", "file": "mit_franka/video_000119.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/42260c9bf042b8c1562c5181.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000120/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "663b531c-43f0-4f96-af80-083bd06f827b", "file": "mit_franka/video_000120.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0d91ee883d87680b2ed31400.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000121/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "6c38f3c6-1258-4943-86bc-a84168201703", "file": "mit_franka/video_000121.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b6254a22cc745e0f7120befd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000122/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b2b95e51-512a-4a64-8087-c146de3145f8", "file": "mit_franka/video_000122.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/418c4e1c08d9cbac04ac4477.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000123/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "3ff3e7bd-a1c5-42da-979e-5cf52da01326", "file": "mit_franka/video_000123.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7a893e3a67176a69aa4b05f4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000124/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "15b63612-8916-4023-8f13-f7fb29add5df", "file": "mit_franka/video_000124.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5a165a6d72b19e1268b062ed.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000125/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "cee4e79a-6cf1-4981-b500-d68ac80c51a2", "file": "mit_franka/video_000125.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bb123dfaed0703fb74363d9c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000126/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "9961c289-ba40-429b-8d89-1147d31dc746", "file": "mit_franka/video_000126.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6ca44f65b54609bf47add7ad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000127/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "e8ff0eff-9796-4374-9502-94d54ff54835", "file": "mit_franka/video_000127.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0049f18c537236cce233b309.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000128/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "be9ee93f-cc2e-40b2-b99e-9aff261867a5", "file": "mit_franka/video_000128.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d0f4862f7fa19f985556a805.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000129/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8cb48ffc-25d4-4337-a07d-963b24e888bc", "file": "mit_franka/video_000129.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cfdc87037bb3c1481af5b99c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000130/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "f5171b6e-6bc8-47b6-8ffd-d8955acc9fea", "file": "mit_franka/video_000130.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c0aec9e3862f4fe7f28a8599.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000131/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "86ad603d-87ca-4fc9-918d-347e28f9974d", "file": "mit_franka/video_000131.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/52a51142a822e69700971cc0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000132/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "196c6353-4dc2-4af0-b65c-a758e7e45345", "file": "mit_franka/video_000132.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2c24707ea3052389a6c0e629.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000133/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d0fce1a9-ea9e-4013-bcf7-85c9f9255f2d", "file": "mit_franka/video_000133.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5e9767401dd7d4d67ba34dd2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000134/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "03132b75-d5b5-469c-a51d-91d2adcb2e42", "file": "mit_franka/video_000134.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3958e68489fcfc4b9a1b76b0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000135/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ef734a6a-b126-4701-ad2e-3bd7bcc3d892", "file": "mit_franka/video_000135.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/437a0f80d5e098ebcb7fe1d5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000136/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ff20e79d-74a8-422d-86e6-48e3d81a5f7d", "file": "mit_franka/video_000136.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/77cc2c0a915a86bdd1ec5e81.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000137/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "95359920-4e27-4363-a133-2ef7e90de9ff", "file": "mit_franka/video_000137.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5f5a96cbc232e6c97918cb7d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000138/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "41288645-ddd4-46af-9402-d67c21b98227", "file": "mit_franka/video_000138.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/258e6f2dd4f938f0b137993f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000139/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "648758dd-b8fb-4aa0-bb5a-286d1d89aadb", "file": "mit_franka/video_000139.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/55fdf988d754b56713cb0037.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000140/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "7b458019-4662-457a-bc46-5615e9a1af17", "file": "mit_franka/video_000140.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8630f7b406eeb0a5f383be08.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000141/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "12354cbd-8e8e-4be6-80d8-c388fe9e1059", "file": "mit_franka/video_000141.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4bc073caf2fdf83e5f723d79.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000142/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "14b766d5-7a47-4730-ad93-433a92162563", "file": "mit_franka/video_000142.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2ec80d8c37f95668df7c231e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000143/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2e5d1a5a-314f-4931-acfe-8a97816ac4ce", "file": "mit_franka/video_000143.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f2e7820f3e1ebc1db53d9060.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000144/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "95d0a35e-ae23-4a48-bdff-00abf0713b26", "file": "mit_franka/video_000144.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/73aa2a398963012f5f86aa91.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000145/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2d3e8649-ef2a-4e08-97c3-e7df066a7e7e", "file": "mit_franka/video_000145.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b172ee62fa7379adfdb739a9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000146/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "46e9bca6-7817-43fb-914b-edcc12e8010a", "file": "mit_franka/video_000146.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/19dbdfbb30f70e69675a2efd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000147/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "5dba8377-0c73-48c0-89f2-85e16fee39bd", "file": "mit_franka/video_000147.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a8cf96f2def63dc876027a82.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000148/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d08f2b68-4269-4b2b-bd78-c400099f2fae", "file": "mit_franka/video_000148.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bf8ed1095bfbb822ef73a3b4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000149/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "4841ae7f-51c0-44d1-981c-7e22b1765ecb", "file": "mit_franka/video_000149.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/53c84b152fb46b1d5ffb595d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000150/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "9c76fad2-fd74-4fda-9625-6694564dd16e", "file": "mit_franka/video_000150.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ee168782a7878f3965c5c516.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000151/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2a8163c8-03ee-4b29-99a1-c6849c349d13", "file": "mit_franka/video_000151.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ec9d7ce3d2589640f6970e9a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000152/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "cfb656ed-cbff-4438-bf36-7193a2658da4", "file": "mit_franka/video_000152.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0d779a4c63472be0c49d3bac.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000153/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "566afd80-feca-456d-908f-31a64c3569c4", "file": "mit_franka/video_000153.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0811f0e595812b8d5df0805c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000154/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "5aa9a854-1ba8-48d2-94b6-0926f5f8fc5a", "file": "mit_franka/video_000154.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3f316f580b895ceed7921873.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000155/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "f921a32c-2344-4918-8329-3a13ff0d27bf", "file": "mit_franka/video_000155.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3d3330461144450901579493.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000156/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "74d78e31-da98-43c8-809a-0cdd6dbd337d", "file": "mit_franka/video_000156.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a2b30565ccce9c6e98771870.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000157/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "3580a08c-c530-4206-9adf-49696348cf63", "file": "mit_franka/video_000157.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/714b21c25aa36c2257227253.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000158/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "58c49e06-a5b9-4165-9178-d838dbb837b8", "file": "mit_franka/video_000158.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3c0c586e912c49fccee0cddc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000159/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "e28c2548-9b6e-40c6-b813-00dd8cfe63ad", "file": "mit_franka/video_000159.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c7a1c30c9358c8415ee3f168.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000160/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "39d6923e-647d-4007-ba43-3523a12cd0eb", "file": "mit_franka/video_000160.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c0a7229331aa6f48b2a80d75.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000161/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2f8ded58-2c09-4aea-bb90-a74d7b67a151", "file": "mit_franka/video_000161.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1dc06462371d8be6afe617f2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000162/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "22bde85d-a824-42c1-8c11-3ce96dc30b29", "file": "mit_franka/video_000162.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c4b93f42530c139811a3f8d0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000163/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "6d6781e8-d812-4968-8e1f-8ae633df7341", "file": "mit_franka/video_000163.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/74446491dda9de227942d435.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000164/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "5e674c40-7e9b-4263-9a97-f7cdf3652ff6", "file": "mit_franka/video_000164.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d025bbadc248712822029a31.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000165/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8965693b-a82e-44b7-be7b-084d5f46f8f4", "file": "mit_franka/video_000165.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c077b7766db5347894ecaddc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000166/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "c11d0ab2-69f5-40df-8334-38efc65658c8", "file": "mit_franka/video_000166.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/452813b5e62035853fc63b30.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000167/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2da3ac9d-c65a-4b97-af12-4836a9a06c0d", "file": "mit_franka/video_000167.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/96455300770b7fcf27cdc877.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000168/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "f1aceb48-98fe-402a-ac86-6cf4989edc6f", "file": "mit_franka/video_000168.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d405875905853e48957e2fd2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000169/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "aced9fe0-caee-4790-b4b7-e4ef20febf1b", "file": "mit_franka/video_000169.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/88ddc8c392c19c2393bd3a64.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000170/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "23c242da-0437-4224-aeb5-9d2d021d40a6", "file": "mit_franka/video_000170.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5d18ee8d6273be5fc36d3c42.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000171/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "86840c22-3271-4784-88d6-4484e219813f", "file": "mit_franka/video_000171.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f660e59eae13450d9af876f8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000172/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "0c1d3d91-a9f0-423e-af58-34d61b90f7cd", "file": "mit_franka/video_000172.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/80ed6d1cc973ab5fb79eaf27.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000173/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "3a8a20cc-3b5c-4866-94c0-46bd7e642c29", "file": "mit_franka/video_000173.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/32e0d9eceb57d63eb6ea49d4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000174/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "fa85c086-d384-4e53-a2d6-a7286b86e5c7", "file": "mit_franka/video_000174.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5405d40c5e6be2c9834ce6ef.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000175/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "06ff8c90-0541-47c1-8445-c51cf92b665b", "file": "mit_franka/video_000175.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4ad7d98e9dffb3248d9d7230.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000176/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "3b289cc8-0e0d-4431-a9e8-685430eac45b", "file": "mit_franka/video_000176.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2cff3f2385d47bd394dac569.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000177/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "86bfc3b8-38bf-412b-8b9c-95bc73d0cc0c", "file": "mit_franka/video_000177.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0dc9e48874b6a3efbb036b7e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000178/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "48ae3c66-1ca7-4dfa-91ef-0abb06817d2c", "file": "mit_franka/video_000178.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f4cec55bab71a4a13afe2565.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000179/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "860f71b0-1e00-4d38-b2e3-42c68cece12c", "file": "mit_franka/video_000179.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9e32266c14919f75e7d40252.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000180/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d929ea18-4f95-4015-8804-919dd091c626", "file": "mit_franka/video_000180.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fbcee6f166954eb02ec74db3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000181/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "bea540e6-0b6b-493b-a89c-048a6f6e8f68", "file": "mit_franka/video_000181.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6cbfc6d0b4762799d1dc2767.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000182/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "e495eabb-9540-4cf3-a8ed-f850a7c4d979", "file": "mit_franka/video_000182.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6a4c1ba908e9e1165b9055cc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000183/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "3b7ddb9b-ab1b-4088-8d85-41d331d59184", "file": "mit_franka/video_000183.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0500e554dff78a544e0f22e0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000184/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "cf424626-1422-47bd-b676-19a75413d057", "file": "mit_franka/video_000184.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/56977eb47fb6ed3f0c97ad7b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000185/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d0753448-c525-4e1b-a96c-4738772cb980", "file": "mit_franka/video_000185.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c8ef6a59e0b8f559f8a14ec6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000186/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "eea29928-bb0a-4e0a-9e92-7765a60f569b", "file": "mit_franka/video_000186.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7b42da0ec3bce19dd7f03209.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000187/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8d857e6b-d6a6-4799-ac84-57d854b1c353", "file": "mit_franka/video_000187.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/929d40ab55ae52abd516681e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000190/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "e44b42ab-b1cc-44c3-aea0-8361c7cc29fa", "file": "mit_franka/video_000190.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the banana from the blue plate and place it on the green plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b77828fde0a6da0f4a6c7a61.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000191/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the banana from the blue plate and place it on the green plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "eddc195e-30c7-405f-aaf6-1ac3a4e4ff09", "file": "mit_franka/video_000191.mp4", "quality_label": "success", "task_norm": "Pick up the banana from the blue plate and place it on the green plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 157, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/851c46452dc5aa72cbc4cbd1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000192/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b013712a-8ae8-4db6-a783-2208a5495aec", "file": "mit_franka/video_000192.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a651aabbd84818fb477e6b98.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000193/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2f98b96c-ea1d-419f-bda1-e5af330d80f6", "file": "mit_franka/video_000193.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8cc26d98b31575dfbb0a4cfd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000194/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "dfe31e07-ba00-418e-bc78-8fd67141fce2", "file": "mit_franka/video_000194.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e1658bb8e1395290120d16de.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000195/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "70d81e02-e13b-40ca-ba35-0145cc226746", "file": "mit_franka/video_000195.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8447174ec098927c1f5544d5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000196/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "56e82482-c443-4336-8d1c-afed307ce817", "file": "mit_franka/video_000196.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/87fda6de370b6301a419cbe5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000197/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b9a63762-7e22-4194-85a2-93d5666d876c", "file": "mit_franka/video_000197.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d928f1d53d05b582aa3f3d01.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000198/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "70e1fe0f-2dc7-4813-b728-4f7b6f0dddb7", "file": "mit_franka/video_000198.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e71bf028974f5f49a3323ef0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000199/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "7f865ab8-c7c9-47f9-b7a9-a1dcf3bf5f41", "file": "mit_franka/video_000199.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/de9d657d67e76259ab2c1a27.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000200/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d2130229-c90e-4f99-89a8-708ee3d7d40c", "file": "mit_franka/video_000200.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b9583801513df4fafb9e0d8c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000201/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "5f4ce0ee-ac41-4d54-bb09-903f1153c78c", "file": "mit_franka/video_000201.mp4", "quality_label": "success", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7a600269b3d684a78718d131.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000202/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ab343a28-eb35-4e31-99b9-ba5fa8225d79", "file": "mit_franka/video_000202.mp4", "quality_label": "success", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1a8cd1259bea5596a0c3b458.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000203/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "29dc44e8-3ae1-4849-aa0f-8efb7e2c6e04", "file": "mit_franka/video_000203.mp4", "quality_label": "failure", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0855f40ebcc8a87d8eff1957.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000204/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "94a8857c-bbaf-4d02-9582-368991ecc094", "file": "mit_franka/video_000204.mp4", "quality_label": "success", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4c3b67484b6f5f9dea6ce75d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000205/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "67ea3c81-d7e1-4ddb-840e-4f16e09a8da2", "file": "mit_franka/video_000205.mp4", "quality_label": "success", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f32c6696483e20a0bd5e6b0e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000206/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "15f075e4-d639-4a40-949e-199a0e5dafcd", "file": "mit_franka/video_000206.mp4", "quality_label": "success", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c8bb536df9c4f5be6fcb5cde.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000207/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "0e067172-7a09-42b1-a71a-2ff359818496", "file": "mit_franka/video_000207.mp4", "quality_label": "success", "task_norm": "Pick up the mouse and place it right next to the laptop, while avoiding spilling coffee", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 160, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/67a7e33243c4940964234063.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000214/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "cbea2c79-7d49-46c8-87de-cb8eddd5360e", "file": "mit_franka/video_000214.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c7a93e87441bd680e6f4381a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000215/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8c794ff2-6cfe-405e-97d2-a4d57b8be79b", "file": "mit_franka/video_000215.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7232415ed7a64367d7d868d3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000218/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "6711b1bb-ebb5-4aff-b9c0-ecc4a37b9437", "file": "mit_franka/video_000218.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b8df5dcbdc6e312dfbee4c2e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000219/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "5a977a2c-6b2d-4c3d-9266-44c66d77f666", "file": "mit_franka/video_000219.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3aa0f97b384ecb637d4befe5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000220/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "a75c2cac-349d-4d15-93ce-ad2f1f7877bc", "file": "mit_franka/video_000220.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9d2fbbb80fd036bc3984aa6e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000221/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "c3938b4a-0d7a-4df2-82ab-397b201449dd", "file": "mit_franka/video_000221.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d05341995ab357dc20eee8d3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000222/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "5920cd7a-c2b3-47ab-bc88-79b18e1024fb", "file": "mit_franka/video_000222.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/db91be15d9e8270a9fa3d74b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000223/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "07b1ec0d-0aec-4b55-a0be-a3d7428bea47", "file": "mit_franka/video_000223.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/12b0dfeb26ee3fd421ec94b2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000224/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "37cba8cf-f5d7-4be2-80bb-749cd6b40afb", "file": "mit_franka/video_000224.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9670f9d7b9c308e5df1cb038.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000225/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "cfe4bd76-ee6c-4366-8ed4-eb4788426005", "file": "mit_franka/video_000225.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3a9a286dcca6c40c264e6c91.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000226/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "60dd34a2-a44b-4f7b-b97a-9189ec164365", "file": "mit_franka/video_000226.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d87546b33f7825e8ee7e92e5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000227/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "49c291d4-ca32-43b0-ab1d-18971e1cc5ba", "file": "mit_franka/video_000227.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5136eab7c5f0c69ef0ce2a05.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000228/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "4d30a0de-bd76-43fb-ab53-42548d043162", "file": "mit_franka/video_000228.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e11967eb5b33e160e1422624.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000229/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2515bfa2-44b9-4b19-b088-876ff875801d", "file": "mit_franka/video_000229.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2fec9f166753726b3f4817a7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000230/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "f5e111a8-154d-4653-a5f6-3d8748f88fec", "file": "mit_franka/video_000230.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c6290fc3f924e8ff889b2d7f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000231/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ccb4bc2e-7531-4c0a-b007-62967a77628d", "file": "mit_franka/video_000231.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e838d42b4d47de3b70537273.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000232/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "15e57e62-b420-43b2-9700-a2dd2e4bf790", "file": "mit_franka/video_000232.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c732e3e0ac12c6533694ae02.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000233/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "55b8baa3-0054-4abe-acba-b3c66577d4fd", "file": "mit_franka/video_000233.mp4", "quality_label": "failure", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9f4ce3e469a66c7be3a42703.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000234/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "2bc5e41a-f2f9-469e-9c61-edea17686a67", "file": "mit_franka/video_000234.mp4", "quality_label": "success", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/38288578b72cc496250b0d94.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000235/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d4bd0b33-6be2-4276-be13-28861e9ef0ee", "file": "mit_franka/video_000235.mp4", "quality_label": "success", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8d0c7fc70a510c8827bc9464.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000236/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "10ffeb3c-19b2-400a-9112-570fcaa5902a", "file": "mit_franka/video_000236.mp4", "quality_label": "success", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8dfbe5121d43ef5a154f323d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000238/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "f1a636c7-5760-4b3e-bdf4-9bd003f0172d", "file": "mit_franka/video_000238.mp4", "quality_label": "success", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0aa734a9dd55e033f022aebc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000240/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "132839cd-841c-480e-940c-9c22d34dd355", "file": "mit_franka/video_000240.mp4", "quality_label": "success", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pour the pebbles from the cup onto the plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/24099e99fe0117887431dce3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000244/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pour the pebbles from the cup onto the plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b00d51d5-9ad6-45cf-aca8-03df33c632e7", "file": "mit_franka/video_000244.mp4", "quality_label": "success", "task_norm": "Pour the pebbles from the cup onto the plate", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 153, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d53668fc7bd1cad5cae26103.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000246/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "5293dca8-22bd-49e0-bf4a-014b4a87fc42", "file": "mit_franka/video_000246.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a6a5ccab792ec11c7536a8b2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000247/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "92704c22-37e3-47d7-8d98-f9a2660e59e1", "file": "mit_franka/video_000247.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3e43347081c8513ebd882aad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000248/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "caf3847b-517e-4255-84e5-55fb56530f14", "file": "mit_franka/video_000248.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c4b2424f70d2c53fb6fafa54.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000249/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "588e3fa8-039d-444b-bcf0-bd39cecd436c", "file": "mit_franka/video_000249.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/26580546b3e49ad9736ab422.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000250/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "91dcc5c5-1b64-4222-ae15-45ddbff63d64", "file": "mit_franka/video_000250.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8b7c019829cb4c59eb66917e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000251/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "4c9422ed-3f72-4eb6-a7eb-feea7c18e696", "file": "mit_franka/video_000251.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/97807926afc896668ad66cf2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000252/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b428c0c7-80ed-4dd8-8c42-8000bea2fd78", "file": "mit_franka/video_000252.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/46545eec355dab9c311e51fd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000253/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "774e444f-681c-4602-9c1a-c87c59cb9e3f", "file": "mit_franka/video_000253.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4d253d67b7436b1c1ea4dd82.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000254/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "37c316ad-2fbf-47de-a713-bd63d7148a00", "file": "mit_franka/video_000254.mp4", "quality_label": "success", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/20ab7299ba757374dbe463e7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000255/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "362444b2-3b7e-4e0a-b025-23e6071c0944", "file": "mit_franka/video_000255.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c659f9ba2b71317c04a9dbea.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000256/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "fd33b95b-e031-4057-9356-453c63fa4f8d", "file": "mit_franka/video_000256.mp4", "quality_label": "success", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e5b78e19862640e79c9406be.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000257/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "cdbebfe5-2858-49e1-b807-e88935c2f003", "file": "mit_franka/video_000257.mp4", "quality_label": "success", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6b0df54159fbd088f242596b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000258/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "23667b25-131a-44c6-aa31-9413c4a18a7a", "file": "mit_franka/video_000258.mp4", "quality_label": "success", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ae06bdaf9ffc19b43d7d2a9e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000259/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "f941604f-ad46-43ad-914a-2c53eef0bd8c", "file": "mit_franka/video_000259.mp4", "quality_label": "success", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/90d78fe0a76563fd90fa0a0b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000260/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "23c5eeec-16a3-4cbd-9614-ea3ae9a47771", "file": "mit_franka/video_000260.mp4", "quality_label": "success", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pull a tissue from the tissue box", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/72382037050996d7c82495a8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000261/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pull a tissue from the tissue box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "b9d0e32f-df63-49b7-80cf-96feb65677ce", "file": "mit_franka/video_000261.mp4", "quality_label": "failure", "task_norm": "Pull a tissue from the tissue box", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/86f901e15f60d7ac5fb5e3a9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000272/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d4b21264-5347-43db-99a0-1aa67d59dae2", "file": "mit_franka/video_000272.mp4", "quality_label": "failure", "task_norm": "Pick up the spoon and place it inside the cup", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/636cda7938d7029aef72de4c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000273/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "4dd57b74-112b-4241-b941-e1530d3d93b9", "file": "mit_franka/video_000273.mp4", "quality_label": "failure", "task_norm": "Pick up the spoon and place it inside the cup", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ddb4ac46a1ec6a448ad21614.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000274/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "0151e02c-f5c5-43a6-be60-3bc377b6500f", "file": "mit_franka/video_000274.mp4", "quality_label": "failure", "task_norm": "Pick up the spoon and place it inside the cup", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9dfb9c1db36187bf5ca0a91c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000275/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "12c4b30a-4d5c-47cf-99da-aa4cd73433f7", "file": "mit_franka/video_000275.mp4", "quality_label": "success", "task_norm": "Pick up the spoon and place it inside the cup", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/71a5b760baa26925d6722806.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000276/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "95bef17e-ae03-4040-ac10-c15c01e2cce3", "file": "mit_franka/video_000276.mp4", "quality_label": "failure", "task_norm": "Pick up the spoon and place it inside the cup", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Pick up the spoon and place it inside the cup", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e5c5bae4f1a7cc90407c0679.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000278/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spoon and place it inside the cup", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "bd5008fe-b0ff-4d7b-b994-b7435b756829", "file": "mit_franka/video_000278.mp4", "quality_label": "success", "task_norm": "Pick up the spoon and place it inside the cup", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/dca9cb88719a38361a171ed7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000280/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "235d3093-4796-4f84-baa9-ce18d6ff1a49", "file": "mit_franka/video_000280.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9f01e05b635028e6ef70ca5f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000281/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "5b261f71-de3a-4626-bb12-6aa2f69af2f0", "file": "mit_franka/video_000281.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/612cffea073a7cccf6f2e3dc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000282/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "aeeb8a39-2eff-4ce2-95da-b3dcf51a7346", "file": "mit_franka/video_000282.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/780df5e82b93095de5542db8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000284/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "26242dd7-a9a6-4290-bb00-30f206f97876", "file": "mit_franka/video_000284.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4e3a7da83c4601f82619a043.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000286/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "9aa6ff0d-a0c0-4952-9faf-14df41ef387a", "file": "mit_franka/video_000286.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/57fec86464d9d3869e908523.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000287/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ebdcb9c5-a3c8-4de7-bf1c-d5e27472ef6a", "file": "mit_franka/video_000287.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ab3915a2aed9862634ae744c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000288/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "d2d0d2e1-536c-4ea7-888f-fa5745aae2d1", "file": "mit_franka/video_000288.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/69b3d43e7bf4a9f210ef12ec.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000289/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "be2f859c-f6ba-484b-9a72-3a95d41ceb08", "file": "mit_franka/video_000289.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3d93eed3da94339b79c7704c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000290/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "ea0b24b3-a13d-4b1d-9fcb-4641918adde3", "file": "mit_franka/video_000290.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2a974a83cd564e6b6fee7dc1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000291/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "9d084f85-1caf-47eb-86e1-6e04360bbcf5", "file": "mit_franka/video_000291.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5dd97b579eef30f50b992845.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000292/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "36f6efbe-618d-4f8a-9248-dfd418f622b2", "file": "mit_franka/video_000292.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/64e83bf364e9b8e000c74091.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000293/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "13fe84df-e15f-46b5-b912-7469492471cd", "file": "mit_franka/video_000293.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e72fac9a67a94cd35ddcf5dc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000294/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "69bb936b-5edd-4ce4-b0ee-0b8014c39137", "file": "mit_franka/video_000294.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d3257d2afd703e27c5b9161f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000295/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "4411a010-2a24-47ae-a7b8-de61de38e8a0", "file": "mit_franka/video_000295.mp4", "quality_label": "failure", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1ca4e8fa17bd44b9e27b74be.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000296/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "87c70b30-5a99-4411-9add-7790f2d5a297", "file": "mit_franka/video_000296.mp4", "quality_label": "success", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4beffc864b4f0a0d205df597.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000297/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "07ba1444-4ea6-45c7-9b2c-3b74a1fdb49f", "file": "mit_franka/video_000297.mp4", "quality_label": "success", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ea1437d0686506166f2a01e9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000298/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "8d1c9687-085e-40e3-b7ab-e693d917f020", "file": "mit_franka/video_000298.mp4", "quality_label": "success", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/25ae0ebdbab12845d3a1b13e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000299/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "03249d25-4221-4b43-bc21-c72bb3846592", "file": "mit_franka/video_000299.mp4", "quality_label": "success", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/80f6e3eda8135eca98435b88.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000300/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "1d9930d7-eb19-4d96-96ba-f3fdcdb2fe72", "file": "mit_franka/video_000300.mp4", "quality_label": "success", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3683bfaa4ded976bc17711d5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000301/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "6d5c218b-66cd-4395-94c5-4024cde9f93c", "file": "mit_franka/video_000301.mp4", "quality_label": "success", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f0e01e147c1a50a6e93ced53.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000302/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "36585075-3c6d-4f06-b112-a90f84556cb4", "file": "mit_franka/video_000302.mp4", "quality_label": "success", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Pick up the spatula and stir the beans in the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f7d04c699265a3d26f045692.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/mit_franka-video_000303/robometer_traj_success", "group_id": "robometer_ood/mit_franka/Pick up the spatula and stir the beans in the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "mit_franka", "rig": "a Franka Panda robot arm", "native_id": "48279aac-ff7f-462d-9a1d-9ab718f8c2dd", "file": "mit_franka/video_000303.mp4", "quality_label": "success", "task_norm": "Pick up the spatula and stir the beans in the pot", "video_info": {"codec": "h264", "width": 426, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 154, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cc4203dec13fcbf9a2ea0a01.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000000/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "d6b86600-6750-4e21-9892-b67442c97841", "file": "usc_franka/video_000000.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0072dce5a2e0296aa21fd341.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000001/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "fd70a190-7043-4a8e-ac24-c3ea943856e5", "file": "usc_franka/video_000001.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f01f73e81747b54c3dfec63e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000004/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "da0ecbfe-d086-4a01-857f-d0cdb9e30b72", "file": "usc_franka/video_000004.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1fb3d361c7d4d9fdd6c6cc90.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000005/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "d8a0dba2-e4b3-4db3-bce7-60538500224d", "file": "usc_franka/video_000005.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Put the yellow plate in the sink", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/169dd0a1a099a820b606574e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000006/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "94af8ff4-0f77-48fd-9a83-cc35fd060e66", "file": "usc_franka/video_000006.mp4", "quality_label": "failure", "task_norm": "Put the yellow plate in the sink", "video_info": {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the yellow plate in the sink", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/507c44d2bb43eaad67947b5d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000007/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "c2a93b8e-5c54-4472-8ed9-d7abee1338fd", "file": "usc_franka/video_000007.mp4", "quality_label": "failure", "task_norm": "Put the yellow plate in the sink", "video_info": {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the yellow plate in the sink", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0b704cd3b1f13c7e609a97d4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000010/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "4a7082b7-018e-49cd-8484-ab15695a71e9", "file": "usc_franka/video_000010.mp4", "quality_label": "success", "task_norm": "Put the yellow plate in the sink", "video_info": {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put banana on plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0b460b0a2cb424ffe89d3edf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000011/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "c006a720-9007-4088-bd91-7b570c1bb238", "file": "usc_franka/video_000011.mp4", "quality_label": "failure", "task_norm": "Put banana on plate", "video_info": {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Put the yellow plate in the sink", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a3ff5fb15da02c4731ec5e03.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000012/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Put the yellow plate in the sink", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "f86a35b5-ab99-4a53-baed-78f83995ef74", "file": "usc_franka/video_000012.mp4", "quality_label": "success", "task_norm": "Put the yellow plate in the sink", "video_info": {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put banana on plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0c3ce8d98b888d0a83da85e0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000013/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "a2dae243-c742-4d25-a09d-d18b6c0baebe", "file": "usc_franka/video_000013.mp4", "quality_label": "failure", "task_norm": "Put banana on plate", "video_info": {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Put banana on plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f3a9410d768e4e46f6329ead.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000016/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "5ff4a784-0bf9-42cc-9e69-66548ca8b1e8", "file": "usc_franka/video_000016.mp4", "quality_label": "success", "task_norm": "Put banana on plate", "video_info": {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Put banana on plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0effdd669890be15a200aedf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000017/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Put banana on plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "115e12f1-8e7a-429c-bc7a-6e4086649d5d", "file": "usc_franka/video_000017.mp4", "quality_label": "success", "task_norm": "Put banana on plate", "video_info": {"codec": "h264", "width": 240, "height": 320, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Stack the red cube on the gray cube", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f306027f4950e3f33f8108a0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000018/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "4804b50b-9efa-4844-8e4d-533bd403f71e", "file": "usc_franka/video_000018.mp4", "quality_label": "failure", "task_norm": "Stack the red cube on the gray cube", "video_info": {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Stack the red cube on the gray cube", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3375e5c9332febf043d2623f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000019/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "a4ed8e1b-48f5-4125-9972-c07674af249e", "file": "usc_franka/video_000019.mp4", "quality_label": "failure", "task_norm": "Stack the red cube on the gray cube", "video_info": {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Stack the red cube on the gray cube", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/580347ff7bc61f52a8906aab.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000022/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "b9eb4443-1d51-4376-822a-aa3492642d04", "file": "usc_franka/video_000022.mp4", "quality_label": "success", "task_norm": "Stack the red cube on the gray cube", "video_info": {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Stack the red cube on the gray cube", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e5048e74b351be631af8deec.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_franka-video_000023/robometer_traj_success", "group_id": "robometer_ood/usc_franka/Stack the red cube on the gray cube", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_franka", "rig": "a Franka Panda robot arm", "native_id": "f8d7a3e1-c44f-42c8-836c-f650283551e9", "file": "usc_franka/video_000023.mp4", "quality_label": "success", "task_norm": "Stack the red cube on the gray cube", "video_info": {"codec": "h264", "width": 240, "height": 426, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/aadd61dadee3f6f1c3d46da1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000000/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "825b35e5-08a7-4f2d-9a57-f2d7a5d3457a", "file": "usc_koch/video_000000.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e539d046a6d3089883fc8523.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000001/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "099d39c2-c632-4b33-bdb4-2ab2b870b7b9", "file": "usc_koch/video_000001.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0a008df3bdb5751e6d1b3923.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000002/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "848b7b3f-c3cd-4814-a185-34216025cb90", "file": "usc_koch/video_000002.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d319f56706d5f687cc3e13f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000003/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "997e4ad9-3b89-4254-86e8-7524534ac12e", "file": "usc_koch/video_000003.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/92d55fe186c736f4e935e923.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000004/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "5ed9b84b-e9f9-4557-ae59-e805128720f6", "file": "usc_koch/video_000004.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/00f490892621e59be43fed0b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000005/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "24f819f4-0a30-4e20-9573-d17e5ef2fc11", "file": "usc_koch/video_000005.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 21}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6256c96f60195bc0d17f5abf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000006/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "0d368af7-daf1-411e-89d1-132d17ec73f5", "file": "usc_koch/video_000006.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3427136fafbefdddd8748b01.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000007/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "1821b533-7108-4abf-9faa-bd68a0698ee4", "file": "usc_koch/video_000007.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/774de4523c7e0b2ab9c42dc7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000008/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "9297e8f6-4440-40ac-8179-441b4d49c483", "file": "usc_koch/video_000008.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/21075a9cf308415285685c4f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000009/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "700cbb25-27e3-4b65-ad81-502204b34071", "file": "usc_koch/video_000009.mp4", "quality_label": "failure", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0ea8c376475214ad385cad07.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000010/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "474ce398-376a-4781-a953-c15b4cc9c865", "file": "usc_koch/video_000010.mp4", "quality_label": "failure", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/4bfc2ef32f3f924e0e68d891.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000011/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "9c1ce8e5-06eb-48ae-9a2a-fedf04db52d6", "file": "usc_koch/video_000011.mp4", "quality_label": "failure", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b867953b8dab43632565130e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000012/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "0687805b-46f9-40d0-b637-a58fe12db327", "file": "usc_koch/video_000012.mp4", "quality_label": "failure", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/40eaeb6ccd25447efe5e4926.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000013/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "cbc83d7d-c80b-46ad-9d78-5991291a78a2", "file": "usc_koch/video_000013.mp4", "quality_label": "failure", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8393e9044843eb9cf143b93f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000014/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "a65d04c6-cef3-40b8-97a1-f811dfabd609", "file": "usc_koch/video_000014.mp4", "quality_label": "failure", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a4ab87bcccb9e7311952bc75.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000015/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "82e7242b-bfcc-416e-be0c-6e46ee0becb5", "file": "usc_koch/video_000015.mp4", "quality_label": "failure", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f7dfa774f8f50e78bd927355.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000016/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "d26a7326-12bd-444e-a621-ec8736184397", "file": "usc_koch/video_000016.mp4", "quality_label": "failure", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2ef32aba1377a43a294e4d3d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000017/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "5a357cb6-7e5b-4dae-971f-c8a037611468", "file": "usc_koch/video_000017.mp4", "quality_label": "failure", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e09318bd6694a36675ea7bc3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000018/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "1407d3b3-1d88-4df2-b8de-8cfbe0c5f314", "file": "usc_koch/video_000018.mp4", "quality_label": "failure", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2659bf1b87a3ad112dc8cff4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000019/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "e362ec6b-f1d9-4a00-9a4a-36e783f97dd2", "file": "usc_koch/video_000019.mp4", "quality_label": "failure", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/68f9de7f8c87059299c0d928.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000020/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "06dd6133-b5d2-4a67-a489-082cf4b4a188", "file": "usc_koch/video_000020.mp4", "quality_label": "failure", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/368d5d07602b6b6913115f72.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000021/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "7f68f4e1-949a-44fd-ace0-d2fc23cff087", "file": "usc_koch/video_000021.mp4", "quality_label": "failure", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e632789105d68aecbe10c9fd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000022/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "a21e3d03-514c-4074-9de6-808b45be72ac", "file": "usc_koch/video_000022.mp4", "quality_label": "failure", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/d8216dd71cdfbc33221a275d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000023/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "0aed6592-49f7-48c0-96ad-dc7e9c7e7ad1", "file": "usc_koch/video_000023.mp4", "quality_label": "failure", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bc4bc2581aa6503b3ef26e19.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000024/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "fbead96c-1271-4b3c-b77d-f48a21b2a0bd", "file": "usc_koch/video_000024.mp4", "quality_label": "failure", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/18e7f9428a91d37fc762d3c0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000025/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "2ee340e3-bc00-4860-b6bf-b23b96fd0359", "file": "usc_koch/video_000025.mp4", "quality_label": "failure", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c8ee74ba64324c30210896e1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000026/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "a9adf6ac-ad21-4ee8-93aa-f8cd609daa6a", "file": "usc_koch/video_000026.mp4", "quality_label": "failure", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f60c76083e6a60ec289d09e9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000027/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "98b482ba-5984-4891-b068-69207b5896ee", "file": "usc_koch/video_000027.mp4", "quality_label": "failure", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/754f9d217b3fb8d2794507bf.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000028/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "135694d3-39d1-4e45-9949-8fd75fc635c5", "file": "usc_koch/video_000028.mp4", "quality_label": "failure", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/af1a51bb4815076f2f07fc48.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000029/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "9dbf386f-4f8f-478c-a1b7-c82d5824ecbc", "file": "usc_koch/video_000029.mp4", "quality_label": "failure", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ced8a313f5e0a7b70f773197.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000030/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "1fe53487-ca8b-403a-b429-36a8d1250447", "file": "usc_koch/video_000030.mp4", "quality_label": "failure", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ea34be0b5fdbf81915d1067d.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000031/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "b7071bb3-99e1-4af1-82f6-c19bbea69911", "file": "usc_koch/video_000031.mp4", "quality_label": "failure", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fac15c4126282488f6bad150.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000032/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "0e417347-7f7d-41ff-89d1-000a52ed164c", "file": "usc_koch/video_000032.mp4", "quality_label": "failure", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f3c3c04f402b7edee41a062c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000033/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "eb59f095-c125-473d-b305-22462d7b66cb", "file": "usc_koch/video_000033.mp4", "quality_label": "failure", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0b2dc3f03ab62ae326a6ed90.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000034/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "504099e1-3595-4b81-b8ff-99bfc6e68b88", "file": "usc_koch/video_000034.mp4", "quality_label": "failure", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2a8399ad3e9d13dbdb4beda5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000035/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "ea21ecd7-794b-48a3-a79b-983ed6d562d9", "file": "usc_koch/video_000035.mp4", "quality_label": "failure", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0d04647b916686027a7e9f26.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000036/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "e370fed8-781f-4fb9-87f0-72f2cd492876", "file": "usc_koch/video_000036.mp4", "quality_label": "failure", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/623383348dff7417b38284d8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000037/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "b979938e-e8ac-4342-b3df-308bf7cdc2d7", "file": "usc_koch/video_000037.mp4", "quality_label": "failure", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/417c6cc6f78ee2802d6357cd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000038/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "83f1b266-7b3a-4104-8122-2f9e99ab60e2", "file": "usc_koch/video_000038.mp4", "quality_label": "failure", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/50d02b44801a93d4e8729e87.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000039/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "84484d2d-5cfa-4e80-9e4c-fd9657fb5bb1", "file": "usc_koch/video_000039.mp4", "quality_label": "failure", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a11cd67a62a7f5badc81c611.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000040/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "ae9e0f4c-57e4-4b1a-affe-127dca818414", "file": "usc_koch/video_000040.mp4", "quality_label": "failure", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/62afe872fcce093fc7306595.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000041/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "38c66aba-3e7f-4143-9d42-cafd96a81488", "file": "usc_koch/video_000041.mp4", "quality_label": "failure", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/48d79577d07a6b32264f6ed3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000042/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "f3e2bbe5-64df-4ccd-9251-8bdb0bf8e195", "file": "usc_koch/video_000042.mp4", "quality_label": "failure", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/da8abc6a5d43d6e56477f2ce.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000043/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "8b16a94c-da02-41e3-855a-683319ac1337", "file": "usc_koch/video_000043.mp4", "quality_label": "failure", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1dc173a57909407873fa700a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000044/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "b2c8a625-3de1-4da2-a704-b50ca0cbd6af", "file": "usc_koch/video_000044.mp4", "quality_label": "failure", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7fad72f2860e8be2327ba98a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000045/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "d56da746-11bc-4c11-8b3d-0f8d4497e24f", "file": "usc_koch/video_000045.mp4", "quality_label": "failure", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2c5fb4347f49cd8002161422.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000046/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "aed484f3-5e85-49e1-bd49-7eadce0fdb29", "file": "usc_koch/video_000046.mp4", "quality_label": "failure", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/dd6cd2bc2f71f46571c17d37.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000047/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "65aa1d59-fd81-4609-9b05-1f09bd66bb47", "file": "usc_koch/video_000047.mp4", "quality_label": "failure", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bdbe81437a702f595995e68b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000048/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "ada56406-a1c2-4228-bfc7-0f2d6c877ed7", "file": "usc_koch/video_000048.mp4", "quality_label": "failure", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/11cb5ad61fd868a3f01eb451.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000049/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "45d8c456-0ae2-4a23-9d5a-f9a946806d5b", "file": "usc_koch/video_000049.mp4", "quality_label": "failure", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0956462da3a8deee366a0fbe.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000050/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "37541aa9-a26c-4a0b-b415-7ce1851e45fc", "file": "usc_koch/video_000050.mp4", "quality_label": "success", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/361ec073714e8a52a48dc6a9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000051/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "cca1425f-b52a-43e3-b8c9-ab2f61ce9499", "file": "usc_koch/video_000051.mp4", "quality_label": "success", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7bc219471d970b672f96d637.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000052/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "1fa7f2e5-548a-410e-af82-1e2f11dbb0d5", "file": "usc_koch/video_000052.mp4", "quality_label": "success", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/50a5c1674d4331d9ddc8a9b4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000053/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "dcaf97d1-1799-4110-9dc1-daeca882e461", "file": "usc_koch/video_000053.mp4", "quality_label": "success", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from left to right", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fc64ee303784fc1ba340ad6e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000054/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from left to right", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "4136d3de-2fa3-449a-a9c4-d9a9206ded09", "file": "usc_koch/video_000054.mp4", "quality_label": "success", "task_norm": "Move the orange cup from left to right", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c131a61907d66872389aa607.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000055/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "46a72685-0ed3-43d9-9154-88d8c4d90c7d", "file": "usc_koch/video_000055.mp4", "quality_label": "success", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2eca56fc9e349a0e5e0af584.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000056/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "1271eb80-7909-42e8-8a04-52c839be6a44", "file": "usc_koch/video_000056.mp4", "quality_label": "success", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/08472d3702906321d46f3728.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000057/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "7af3dd22-3db0-4426-938b-47ec12c1e667", "file": "usc_koch/video_000057.mp4", "quality_label": "success", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/36547d6e5fe1b18d5eb9b507.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000058/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "cce54bfa-775a-446e-bad7-4ebd0d4e6428", "file": "usc_koch/video_000058.mp4", "quality_label": "success", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Move the orange cup from right to left", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/27a83fae7cee812682e800ae.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000059/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Move the orange cup from right to left", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "7e6acdf7-1393-4cd0-8de1-cc4fe7590380", "file": "usc_koch/video_000059.mp4", "quality_label": "success", "task_norm": "Move the orange cup from right to left", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1cee21b2ba5c24d44d609971.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000060/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "b1224526-a8c1-4bc8-ba54-f80e5f6aa33c", "file": "usc_koch/video_000060.mp4", "quality_label": "success", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c7672da87e1edbbcbcfb9aad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000061/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "bade4079-d263-4f66-afae-22c87b06a3e9", "file": "usc_koch/video_000061.mp4", "quality_label": "success", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2076fad185f6cf6aa8944a69.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000062/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "a4c50b45-4262-491b-9333-aa2628454c58", "file": "usc_koch/video_000062.mp4", "quality_label": "success", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fdfba1a57177e17896ce7044.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000063/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "1b743da9-8a11-4058-8dff-01f92e774ca9", "file": "usc_koch/video_000063.mp4", "quality_label": "success", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the blue trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/68ca2c772fc944b62d2042bd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000064/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the blue trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "7d6947fd-1a74-4d77-852e-c2c9b465b95a", "file": "usc_koch/video_000064.mp4", "quality_label": "success", "task_norm": "Open the blue trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f8d68ad43f3091cfc95ae819.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000065/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "a3571be2-f4a8-4ea3-8f54-1fa4733e6e73", "file": "usc_koch/video_000065.mp4", "quality_label": "success", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9e5cbe69b8763b900b1db33e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000066/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "e517076b-2ed5-4da3-89d0-4a5b2e1d84fa", "file": "usc_koch/video_000066.mp4", "quality_label": "success", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2375bfa823288c496727b5aa.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000067/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "f83e7747-d87d-450f-9bda-c943c0a40717", "file": "usc_koch/video_000067.mp4", "quality_label": "success", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/eec326ce75370b88d0b5cad8.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000068/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "661419ab-d66a-4025-8b88-ed6891cc58a9", "file": "usc_koch/video_000068.mp4", "quality_label": "success", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the green trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c39a245e61b7512fcbd1d0b0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000069/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the green trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "c3589188-8b59-462e-ba1e-23183b5608da", "file": "usc_koch/video_000069.mp4", "quality_label": "success", "task_norm": "Open the green trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/012fe793785bd2f01425fac5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000070/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "0f6a53ad-2490-4687-a2d6-6478ad374733", "file": "usc_koch/video_000070.mp4", "quality_label": "success", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/acd01ad9d0ee4734ecc00867.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000071/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "cf7f8d89-b361-47fe-a806-126bf4d62406", "file": "usc_koch/video_000071.mp4", "quality_label": "success", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8edd63615a9a2d341073dcfb.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000072/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "29f4a914-5bec-4cf9-b8e0-05fd611fc207", "file": "usc_koch/video_000072.mp4", "quality_label": "success", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/cb053787c218021cda06a9f1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000073/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "5d703190-0df1-4daa-9088-0a45852f696c", "file": "usc_koch/video_000073.mp4", "quality_label": "success", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Open the red trash bin", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a6d6f633615b3e9dc9852020.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000074/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Open the red trash bin", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "1b1f62db-8d77-437e-875a-c2d622fe36bd", "file": "usc_koch/video_000074.mp4", "quality_label": "success", "task_norm": "Open the red trash bin", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1b5497c6d672b1215ed8ab2e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000075/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "8514edb2-eba2-47f0-bafe-8eba4e21581e", "file": "usc_koch/video_000075.mp4", "quality_label": "success", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/67d3247305e19238750ac1fc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000076/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "d7d5bb5c-84a3-4ca5-945a-d3178f9c0f34", "file": "usc_koch/video_000076.mp4", "quality_label": "success", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c8c6f3842918bfe48f7163f7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000077/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "678b2d21-3e27-41a4-88a6-0d8a39d58a5e", "file": "usc_koch/video_000077.mp4", "quality_label": "success", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ed24f6714f3fba005ac574cc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000078/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "e69006dd-9929-4778-9ac8-772778533dcf", "file": "usc_koch/video_000078.mp4", "quality_label": "success", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5f62084205c68a23f829e8b7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000079/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "4d324a24-2ef1-4db0-b336-cca23179abee", "file": "usc_koch/video_000079.mp4", "quality_label": "success", "task_norm": "Separate the purple and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1d56a18fd02ea151842e85b0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000080/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "6b399eea-7a64-47ee-acb4-cce20dcea553", "file": "usc_koch/video_000080.mp4", "quality_label": "success", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/726df9d19ea580f8d96f924e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000081/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "45d64f1b-a29c-4319-9589-aeb65ed8a153", "file": "usc_koch/video_000081.mp4", "quality_label": "success", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c10fa8f778c2cea6498fd21b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000082/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "46cd47bb-f690-4937-96a3-124567a31bb3", "file": "usc_koch/video_000082.mp4", "quality_label": "success", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0d18b1acd5e5e65efbb3a687.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000083/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "de2eaf57-eb9d-4b9c-a02a-9042cfc89c3f", "file": "usc_koch/video_000083.mp4", "quality_label": "success", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the purple and red cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/00c8dc0856e5e04ce2b7ba73.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000084/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the purple and red cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "dd8c85b1-a78d-4427-89e7-13389a2452af", "file": "usc_koch/video_000084.mp4", "quality_label": "success", "task_norm": "Separate the purple and red cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/481022d159e4dcc05f126069.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000085/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "7b5064ae-99eb-40ac-92e1-461eaf68f42f", "file": "usc_koch/video_000085.mp4", "quality_label": "success", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/57a3da9999c6a617d40dd6a5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000086/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "4737d874-045b-4596-aa0d-8841baca7581", "file": "usc_koch/video_000086.mp4", "quality_label": "success", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9b929b7edf15ea22f6fdd837.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000087/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "a80e0d43-99e8-432f-ac92-865377371225", "file": "usc_koch/video_000087.mp4", "quality_label": "success", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/73d5dcd8fe55da89715d5452.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000088/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "a670ad76-0983-4b9a-b59a-6d4f76bb29fa", "file": "usc_koch/video_000088.mp4", "quality_label": "success", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Separate the red and orange cups", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/358971561d8c777b17300e95.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000089/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Separate the red and orange cups", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "6789878d-b12a-4666-a897-329f07d794c8", "file": "usc_koch/video_000089.mp4", "quality_label": "success", "task_norm": "Separate the red and orange cups", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bff83d49038831fd47cd5e8f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000140/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "40073104-3e83-46ad-9af1-0c3bcf2ca384", "file": "usc_koch/video_000140.mp4", "quality_label": "success", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/602d3a7ee2de086472e14717.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000141/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "5cfc02b2-5402-44d7-9314-d490dfb48f10", "file": "usc_koch/video_000141.mp4", "quality_label": "success", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/af2d593574521dbbfc7a65b2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000142/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "61ce528d-4829-4878-9647-0498d3e2b6b9", "file": "usc_koch/video_000142.mp4", "quality_label": "success", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2289bea937981ca481a9f7a3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000143/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "414634c3-c744-42f6-a9b3-249aef0316fa", "file": "usc_koch/video_000143.mp4", "quality_label": "success", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the black marker away in the blue trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fca8a4488a2a44c08df29ee9.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000144/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the black marker away in the blue trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "36925fed-6878-4520-beac-0490de9cc03c", "file": "usc_koch/video_000144.mp4", "quality_label": "success", "task_norm": "Throw the black marker away in the blue trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b1dcc009840151fb1ebd8596.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000145/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "b1f1ebcd-adc3-42f8-9a3b-f393f473b269", "file": "usc_koch/video_000145.mp4", "quality_label": "success", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2b061e39d8bd94c2898cd0c4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000146/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "f30457a0-5cb7-4412-b873-e63d93de6ad1", "file": "usc_koch/video_000146.mp4", "quality_label": "success", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7e22bb51ee3ec28b63622914.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000147/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "1705cd6f-552d-4250-8e49-0f708c092b50", "file": "usc_koch/video_000147.mp4", "quality_label": "success", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/09309f435866a485908b8520.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000148/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "def91146-877c-4277-83c1-741029f542bf", "file": "usc_koch/video_000148.mp4", "quality_label": "success", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Throw the orange cup away in the red trash can", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a5cb584e04b19b0250b44e10.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_koch-video_000149/robometer_traj_success", "group_id": "robometer_ood/usc_koch/Throw the orange cup away in the red trash can", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_koch", "rig": "a Koch low-cost robot arm", "native_id": "3faeca25-e362-4701-933e-631858a0d3b7", "file": "usc_koch/video_000149.mp4", "quality_label": "success", "task_norm": "Throw the orange cup away in the red trash can", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 32}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 152, "role": "outcome"}}
{"state": {"task": "Open the bottle", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a3f190e0ac5c9b70684f7ebd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000000/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the bottle", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "ec723696-61c2-478a-a21f-0c4bf7f9a0d7", "file": "usc_trossen/video_000000.mp4", "quality_label": "success", "task_norm": "Open the bottle", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Open the bottle", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9a5b9064f94e277a4f6361de.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000001/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the bottle", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "c243382e-cb06-42fb-a1ce-ec787a7da218", "file": "usc_trossen/video_000001.mp4", "quality_label": "success", "task_norm": "Open the bottle", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Open the bottle", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f98edd1610c74695b0ef170a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000002/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the bottle", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "bfad213d-ef98-41ea-9023-174e018287a0", "file": "usc_trossen/video_000002.mp4", "quality_label": "failure", "task_norm": "Open the bottle", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Open the bottle", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1b5134c61cf5fe4fba1f7768.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000003/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the bottle", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "cc2a5e47-6f33-4763-9f7c-ece094b42225", "file": "usc_trossen/video_000003.mp4", "quality_label": "success", "task_norm": "Open the bottle", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Open the bottle", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/51a9b9b331dc4f29467359c1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000004/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the bottle", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "da5b17da-f053-47f0-81d4-275cd1bd668b", "file": "usc_trossen/video_000004.mp4", "quality_label": "failure", "task_norm": "Open the bottle", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Open the red drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9e35c22104c1dd795fbf8e5e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000005/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "fab793f7-2b04-4425-8af9-9e50ea6e0079", "file": "usc_trossen/video_000005.mp4", "quality_label": "success", "task_norm": "Open the red drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Open the red drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/2f84958616ac4373bda9d3fc.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000006/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "6eeb8f4c-2929-4b6f-a85a-2038b6302784", "file": "usc_trossen/video_000006.mp4", "quality_label": "success", "task_norm": "Open the red drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Open the red drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ae7ef8156d8ea6c2984b11f1.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000007/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "c8613f0d-474c-46ec-958d-f3bcfb132372", "file": "usc_trossen/video_000007.mp4", "quality_label": "success", "task_norm": "Open the red drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Open the red drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/595ac1286fff89c431d696af.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000008/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "4798c411-caa3-405e-9afa-132d0a2617f1", "file": "usc_trossen/video_000008.mp4", "quality_label": "failure", "task_norm": "Open the red drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Open the red drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/8a7cb858a552adc296e98610.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000009/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Open the red drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "fcdd7b74-542e-455f-aca8-f6bd110144a8", "file": "usc_trossen/video_000009.mp4", "quality_label": "failure", "task_norm": "Open the red drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Remove the lid from the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/32d1cdb85ab9e2250d8dc454.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000013/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Remove the lid from the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "8d2311da-1e57-4117-b6fe-52332558902b", "file": "usc_trossen/video_000013.mp4", "quality_label": "success", "task_norm": "Remove the lid from the pot", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 148, "role": "outcome"}}
{"state": {"task": "Remove the lid from the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/3262adb92cb76178a6f730bb.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000014/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Remove the lid from the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "08dec425-2969-4989-9c40-8468babfad1d", "file": "usc_trossen/video_000014.mp4", "quality_label": "success", "task_norm": "Remove the lid from the pot", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 148, "role": "outcome"}}
{"state": {"task": "Unzip the pencil case", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/72d1ef3a1d61c4171950e017.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000016/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "043753c3-f59f-494f-944f-52fdd604f7b2", "file": "usc_trossen/video_000016.mp4", "quality_label": "success", "task_norm": "Unzip the pencil case", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Stir the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7deed021a20a2b57fc9d7ef4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000017/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Stir the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "92d0676b-eb5c-4f95-961c-d7320be8c3ca", "file": "usc_trossen/video_000017.mp4", "quality_label": "success", "task_norm": "Stir the pot", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Uncap the red pen", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b7778bd695f292664c188dbd.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000018/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Uncap the red pen", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "2617cb6b-180f-4ffb-b9e7-d87dcf914de0", "file": "usc_trossen/video_000018.mp4", "quality_label": "success", "task_norm": "Uncap the red pen", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Unzip the pencil case", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5c8ae78c938b81ece001df70.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000019/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "bf3d54f4-482e-4f3a-a20f-bbcdc8c699e7", "file": "usc_trossen/video_000019.mp4", "quality_label": "failure", "task_norm": "Unzip the pencil case", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Stir the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/bf421eb7f4c17b9e61117e8c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000020/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Stir the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "ababf79a-774d-4e0e-930f-24c7e0cf8a51", "file": "usc_trossen/video_000020.mp4", "quality_label": "success", "task_norm": "Stir the pot", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Stir the pot", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/93e8aec4e0eb06bf5f965fe6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000021/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Stir the pot", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "8bc4e6db-3dfb-4d9a-88d5-a7790ea4a9e7", "file": "usc_trossen/video_000021.mp4", "quality_label": "success", "task_norm": "Stir the pot", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 146, "role": "outcome"}}
{"state": {"task": "Unzip the pencil case", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/762302281e04f3e3ddf8bec5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000022/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "1372e792-8b26-41a5-aa02-01528fc941a2", "file": "usc_trossen/video_000022.mp4", "quality_label": "failure", "task_norm": "Unzip the pencil case", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Unzip the pencil case", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/111ef9aec7f3a4c7bb582ca0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000024/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "e8b107bc-985d-4093-85a9-fa3e4a88eeaa", "file": "usc_trossen/video_000024.mp4", "quality_label": "success", "task_norm": "Unzip the pencil case", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Unzip the pencil case", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9e676400acb0c58b82582335.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_trossen-video_000026/robometer_traj_success", "group_id": "robometer_ood/usc_trossen/Unzip the pencil case", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_trossen", "rig": "a Trossen bimanual (ALOHA-style) robot", "native_id": "cd11463a-39eb-45ce-9bb9-0b9f76e4f63d", "file": "usc_trossen/video_000026.mp4", "quality_label": "success", "task_norm": "Unzip the pencil case", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 30.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Close the drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/159c61767381c90e054cdd51.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000000/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "9f99930c-cbf8-46b0-882f-ff7b3fcc4e93", "file": "usc_xarm/video_000000.mp4", "quality_label": "failure", "task_norm": "Close the drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Close the drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ac6e7259397b2dda9bfefd1c.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000001/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "8d1d7235-b515-47b8-b8e8-c8d913d85ab7", "file": "usc_xarm/video_000001.mp4", "quality_label": "failure", "task_norm": "Close the drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Close the drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e7c46f757a9f94613489f630.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000004/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "4f6262d5-427f-4fb1-9e88-0bf7c319e434", "file": "usc_xarm/video_000004.mp4", "quality_label": "success", "task_norm": "Close the drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Close the drawer", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/aebe876dceec2ba599c6e1c6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000005/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Close the drawer", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "8ddf2911-58ad-46ba-87cd-7710f2920fbf", "file": "usc_xarm/video_000005.mp4", "quality_label": "success", "task_norm": "Close the drawer", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c791d5e90a895046d4c48ad3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000006/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "887bb1d0-ec5a-4f9d-844f-56f95a787790", "file": "usc_xarm/video_000006.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a5c65de34c0aac6903353ba4.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000007/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "73b0fb7d-dc98-4a97-a39e-fe99acac7b2c", "file": "usc_xarm/video_000007.mp4", "quality_label": "failure", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e60d984ae0bbb741237ae44b.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000010/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "e5b8e128-8655-4f86-b68c-0e5d8647c68e", "file": "usc_xarm/video_000010.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Fold the towel in half", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6de0729023c25daffa3f9577.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000011/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Fold the towel in half", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "71847ff7-3199-4794-bc44-373cc5315777", "file": "usc_xarm/video_000011.mp4", "quality_label": "success", "task_norm": "Fold the towel in half", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Put blue cup in sink", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/fad73ee6d8815d2d11b8e41f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000012/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "0bd8cd2f-85b2-4143-992a-2e185c502103", "file": "usc_xarm/video_000012.mp4", "quality_label": "failure", "task_norm": "Put blue cup in sink", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Put blue cup in sink", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5b3a2516c808ba67a8f6c7ff.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000013/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "dab34346-c997-4a7b-b4c3-28f8124d0427", "file": "usc_xarm/video_000013.mp4", "quality_label": "failure", "task_norm": "Put blue cup in sink", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Put blue cup in sink", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5607241f660c236fa1c4ab1f.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000016/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "84b5e7d9-7f36-4148-a635-196c90eb2b26", "file": "usc_xarm/video_000016.mp4", "quality_label": "success", "task_norm": "Put blue cup in sink", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Put blue cup in sink", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f0c3fb6e3f5b02d691dce852.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000017/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put blue cup in sink", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "bbd59642-b65c-4f8d-9ab6-5ae9f5fe73cd", "file": "usc_xarm/video_000017.mp4", "quality_label": "success", "task_norm": "Put blue cup in sink", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 147, "role": "outcome"}}
{"state": {"task": "Put the banana on the red plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7685bfe0990f9426409242c7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000018/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "f7dea129-57fa-4f5d-8134-679626274759", "file": "usc_xarm/video_000018.mp4", "quality_label": "failure", "task_norm": "Put the banana on the red plate", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the banana on the red plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5b5d2f772189aa4f816dfae7.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000019/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "5244aaaa-04a0-4ccd-bbad-29b2e6c8023c", "file": "usc_xarm/video_000019.mp4", "quality_label": "failure", "task_norm": "Put the banana on the red plate", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the banana on the red plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/54358380194193aa8056daf3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000022/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "fb7d331c-0c73-4856-836f-42576128d95b", "file": "usc_xarm/video_000022.mp4", "quality_label": "success", "task_norm": "Put the banana on the red plate", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the banana on the red plate", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/9db2d885d74bd1599a325706.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000023/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put the banana on the red plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "ffc600a7-d58a-40e9-b4c4-52da5596fbf8", "file": "usc_xarm/video_000023.mp4", "quality_label": "success", "task_norm": "Put the banana on the red plate", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the eggplant in the black bowl", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/eada55b4dc588560761fedf2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000024/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "ed87a481-15e0-4cbb-bf2c-05e8bd070c73", "file": "usc_xarm/video_000024.mp4", "quality_label": "failure", "task_norm": "Put the eggplant in the black bowl", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Put the eggplant in the black bowl", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/6fba71d11993ea2c64f81d12.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000025/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "0947d5d3-1346-472e-9640-4e5c4f9141b7", "file": "usc_xarm/video_000025.mp4", "quality_label": "failure", "task_norm": "Put the eggplant in the black bowl", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Put the eggplant in the black bowl", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/15d1796342ba4f840af450db.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000028/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "312f4b8a-6633-43bd-9094-e6ec964d609a", "file": "usc_xarm/video_000028.mp4", "quality_label": "success", "task_norm": "Put the eggplant in the black bowl", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Put the eggplant in the black bowl", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5ca1d638d6b2c79f4be8e684.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000029/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Put the eggplant in the black bowl", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "70cb5268-c3e6-4e87-99a1-95aa4ffe2899", "file": "usc_xarm/video_000029.mp4", "quality_label": "success", "task_norm": "Put the eggplant in the black bowl", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Stack the green shape on the red cylinder", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/41dba0e9c4c8d75b0fde3500.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000030/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "64e17159-9d91-4147-bfe6-fe19c037f1b6", "file": "usc_xarm/video_000030.mp4", "quality_label": "failure", "task_norm": "Stack the green shape on the red cylinder", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Stack the green shape on the red cylinder", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/13841d846ef9da3a364a595e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000031/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "f61a687b-c4d2-4bb8-9bc6-22d922c891e7", "file": "usc_xarm/video_000031.mp4", "quality_label": "failure", "task_norm": "Stack the green shape on the red cylinder", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Stack the green shape on the red cylinder", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/5bb05087790941c856429561.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000034/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "96fbffd6-97ce-464c-a63e-75f96c745592", "file": "usc_xarm/video_000034.mp4", "quality_label": "success", "task_norm": "Stack the green shape on the red cylinder", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Stack the green shape on the red cylinder", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c8ff7bfc1f5425bb0e9a9db6.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/usc_xarm-video_000035/robometer_traj_success", "group_id": "robometer_ood/usc_xarm/Stack the green shape on the red cylinder", "family": "robometer_traj_success", "extra": {"sub_benchmark": "usc_xarm", "rig": "a UFactory xArm robot arm", "native_id": "ae38a588-1a6e-433b-8317-fff8fbcee422", "file": "usc_xarm/video_000035.mp4", "quality_label": "success", "task_norm": "Stack the green shape on the red cylinder", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Press the button.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e47ab17188ac86bf6e8948b2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000000/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Press the button", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "c8ec8598-f543-4765-9b0a-f573d0fd0b88", "file": "utd_so101/video_000000.mp4", "quality_label": "failure", "task_norm": "Press the button", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Put the marker into the pen cup.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e7288c0900b91ceab3a27381.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000001/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the marker into the pen cup", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "72c73b9d-bc33-4181-804d-bb3096034da4", "file": "utd_so101/video_000001.mp4", "quality_label": "failure", "task_norm": "Put the marker into the pen cup", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the red bowl on the blue plate.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/667ab634edd7f64eb044dbba.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000002/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the red bowl on the blue plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "a5bdb049-3422-4fbb-bfe7-2c637057a6fd", "file": "utd_so101/video_000002.mp4", "quality_label": "failure", "task_norm": "Put the red bowl on the blue plate", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Put the bread in the oven.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e2fad2460c1dad7372b2de59.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000003/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the bread in the oven", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "83cbb89e-6649-4456-9d93-9a50670e51b3", "file": "utd_so101/video_000003.mp4", "quality_label": "failure", "task_norm": "Put the bread in the oven", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 148, "role": "outcome"}}
{"state": {"task": "Collect the fork to the yellow box.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0abd353607ecf5bdc45effff.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000004/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Collect the fork to the yellow box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "77f40402-ef04-45b1-ae7c-9be90a6d80dc", "file": "utd_so101/video_000004.mp4", "quality_label": "failure", "task_norm": "Collect the fork to the yellow box", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Stack the green block on the red block.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e02d83f69d44eac663f3a41e.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000005/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Stack the green block on the red block", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "805de0a8-f4cf-490d-8e2f-bc6cb70a0130", "file": "utd_so101/video_000005.mp4", "quality_label": "failure", "task_norm": "Stack the green block on the red block", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Put the pear in the yellow plate.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/f242921a67eefbd7d5967ac3.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000006/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the pear in the yellow plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "65d1db25-46ce-4175-8b52-fffb517b49bf", "file": "utd_so101/video_000006.mp4", "quality_label": "failure", "task_norm": "Put the pear in the yellow plate", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Add pepper to the green bowl.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1315ec7b9cedc6721e9e43e2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000007/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Add pepper to the green bowl", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "af70eeb2-e86d-45b7-a5fb-a2793c736072", "file": "utd_so101/video_000007.mp4", "quality_label": "failure", "task_norm": "Add pepper to the green bowl", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 148, "role": "outcome"}}
{"state": {"task": "Put the eraser to the blue pencil box.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/0c5cd5fdf76a249b1ece9a0a.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000010/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the eraser to the blue pencil box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "e838f14d-2c55-45e2-9e00-d8883880e9af", "file": "utd_so101/video_000010.mp4", "quality_label": "failure", "task_norm": "Put the eraser to the blue pencil box", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 151, "role": "outcome"}}
{"state": {"task": "Put the red cup on the purple coaster.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": false, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7a10019f18babcbd72393737.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000011/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the red cup on the purple coaster", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "d5c4164a-06b4-4bf0-871e-bbf6c736190a", "file": "utd_so101/video_000011.mp4", "quality_label": "failure", "task_norm": "Put the red cup on the purple coaster", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Press the button.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/ee9f94f244bca1673fb8afd0.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000019/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Press the button", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "5296a584-bd36-43a3-9b64-58980b7814da", "file": "utd_so101/video_000019.mp4", "quality_label": "success", "task_norm": "Press the button", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 145, "role": "outcome"}}
{"state": {"task": "Put the bread in the oven.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b594e8612baa803f1be647ad.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000021/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the bread in the oven", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "0641becc-c8bb-4c92-a56e-8dbd946f465e", "file": "utd_so101/video_000021.mp4", "quality_label": "success", "task_norm": "Put the bread in the oven", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 148, "role": "outcome"}}
{"state": {"task": "Put the red bowl on the blue plate.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/c3f88849f49cf7b8c778aefe.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000022/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the red bowl on the blue plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "e5088127-7e1a-4526-9927-54e2e8f4b5b5", "file": "utd_so101/video_000022.mp4", "quality_label": "success", "task_norm": "Put the red bowl on the blue plate", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Put the marker into the pen cup.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/008a8744f4eac6e3817a4ed2.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000023/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the marker into the pen cup", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "dfb9e90c-75a0-4eb2-83c3-4e38cb101e0e", "file": "utd_so101/video_000023.mp4", "quality_label": "success", "task_norm": "Put the marker into the pen cup", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the pear in the yellow plate.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/1506017b3dfc4d19cf045b17.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000024/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the pear in the yellow plate", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "2a671702-e614-4354-8098-29bc7a6b0068", "file": "utd_so101/video_000024.mp4", "quality_label": "success", "task_norm": "Put the pear in the yellow plate", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Stack the green block on the red block.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/453f1c43077775bfb19c2640.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000025/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Stack the green block on the red block", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "b48899a2-969c-4fa0-9ecb-827456ff935e", "file": "utd_so101/video_000025.mp4", "quality_label": "success", "task_norm": "Stack the green block on the red block", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Collect the fork to the yellow box.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/a522022ac085bbf95256d902.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000026/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Collect the fork to the yellow box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "17115224-7835-4a87-a7e0-f4fd6e7339f5", "file": "utd_so101/video_000026.mp4", "quality_label": "success", "task_norm": "Collect the fork to the yellow box", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 149, "role": "outcome"}}
{"state": {"task": "Put the eraser to the blue pencil box.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/e925b99b06fff2f79db0a3c5.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000027/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the eraser to the blue pencil box", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "3ccee87c-e7db-41c1-9e07-b97110e95b92", "file": "utd_so101/video_000027.mp4", "quality_label": "success", "task_norm": "Put the eraser to the blue pencil box", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 151, "role": "outcome"}}
{"state": {"task": "Add pepper to the green bowl.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/b55c0ae7816a6ca0ab34f964.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000028/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Add pepper to the green bowl", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "6dcbfb2a-a612-433d-a64c-c3083a32337e", "file": "utd_so101/video_000028.mp4", "quality_label": "success", "task_norm": "Add pepper to the green bowl", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 148, "role": "outcome"}}
{"state": {"task": "Put the red cup on the purple coaster.", "video": "the rollout"}, "questions": {"decision": {"type": "noul", "instructions": "You are watching one rollout of a robot attempting a manipulation task. The state gives the task instruction; the attached video is the whole rollout from a fixed camera, from the first to the last frame. Did the robot complete the task by the end of the video?", "label": true, "criteria": {"true": "Success: by the end of the video the task is fully accomplished as instructed (the right objects end in the requested place or state).", "false": "Failure: by the end of the video the task is not accomplished - the robot did not attempt it, attempted but did not finish, or left the objects in a wrong place or state."}, "src": "robometer_traj_success"}}, "media": [{"type": "video", "uri": "media/robometer_ood/7c65a6748fd93d21b3246f54.mp4"}], "_meta": {"source": "robometer_ood", "origin": "robometer/rbm-1m-ood@3f3ee69ecc0ee0db627684da7d599ec26d08175c", "track": "public", "modality": "video", "variant": "clean", "label_source": "human_expert", "license": "Apache-2.0", "id": "robometer_ood/utd_so101-video_000029/robometer_traj_success", "group_id": "robometer_ood/utd_so101/Put the red cup on the purple coaster", "family": "robometer_traj_success", "extra": {"sub_benchmark": "utd_so101", "rig": "an SO-101 robot arm", "native_id": "2cdf5d90-ad7d-495e-865c-bdc33dd74bf5", "file": "utd_so101/video_000029.mp4", "quality_label": "success", "task_norm": "Put the red cup on the purple coaster", "video_info": {"codec": "h264", "width": 320, "height": 240, "fps": 10.0, "frames": 64}, "media_policy": "upstream MP4 copied byte-for-byte (original resolution, frame rate and frames)"}, "n_media": 1, "approx_tokens": 150, "role": "outcome"}}
{"state": {"task": "Place the orange inside the white bowl", "subtask": "Place the orange inside the white bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/46c539bc4afd7676a7f2a036.jpg"}, {"type": "image", "uri": "media/robovqa/37463389fde6652727067df4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0000/robovqa_exec_success", "group_id": "robovqa/0", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 0, "taskvar": "0", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Place the apple inside the white bowl", "subtask": "Place the apple inside the white bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/069e4a3cffaa2483166bc1aa.jpg"}, {"type": "image", "uri": "media/robovqa/ad0f83f9e0566b39e2e776e4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0001/robovqa_exec_success", "group_id": "robovqa/1", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 1, "taskvar": "1", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Place the apple inside the white bowl", "subtask": "Place the apple inside the white bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/cebc1637c09decf296713200.jpg"}, {"type": "image", "uri": "media/robovqa/a240a53017e97a57842480da.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0002/robovqa_exec_success", "group_id": "robovqa/2", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 2, "taskvar": "2", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Place the apple inside the white bowl", "subtask": "Place the apple inside the white bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/65854a81a92a0207df96682e.jpg"}, {"type": "image", "uri": "media/robovqa/1be822e132e55ff7e3c61c5b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0003/robovqa_exec_success", "group_id": "robovqa/3", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 3, "taskvar": "3", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the packet on the table", "subtask": "place the packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5ecea4a175ea1fe6c0acf26e.jpg"}, {"type": "image", "uri": "media/robovqa/503af015dbbf6593cd474ce7.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0004/robovqa_exec_success", "group_id": "robovqa/4", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 4, "taskvar": "4", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the banana in the bowl", "subtask": "place the banana in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/defd0b777b4559af0705efe4.jpg"}, {"type": "image", "uri": "media/robovqa/c110e8d0534db3ea4e560b28.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0005/robovqa_exec_success", "group_id": "robovqa/5", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 5, "taskvar": "5", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the lemon in the bowl", "subtask": "place the lemon in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5743f29f73c4eee874006376.jpg"}, {"type": "image", "uri": "media/robovqa/a41f34c92f8cfdf39ae9a7af.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0006/robovqa_exec_success", "group_id": "robovqa/6", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 6, "taskvar": "6", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the packet on the table", "subtask": "place the packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/e2819d9d15a502fc99275b9d.jpg"}, {"type": "image", "uri": "media/robovqa/30da4c4f95d3de470d3b7217.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0007/robovqa_exec_success", "group_id": "robovqa/7", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 7, "taskvar": "7", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put book on the table", "subtask": "put book on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/719ec286faa339a229bbb077.jpg"}, {"type": "image", "uri": "media/robovqa/e58494ce345ccc664ca4fb49.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0008/robovqa_exec_success", "group_id": "robovqa/8", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 8, "taskvar": "8", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put book on the table", "subtask": "put book on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5a2b04293fd0324e73544eb1.jpg"}, {"type": "image", "uri": "media/robovqa/442d60fbf5af7a9270d1cc1b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0009/robovqa_exec_success", "group_id": "robovqa/9", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 9, "taskvar": "9", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put book on the table", "subtask": "put book on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/0070e1c3eb7021880660f5a2.jpg"}, {"type": "image", "uri": "media/robovqa/596fd613de2f7d83aa4691ff.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0010/robovqa_exec_success", "group_id": "robovqa/10", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 10, "taskvar": "10", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "pick up the sanitizer", "subtask": "pick up the sanitizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/ae841c0dbce0c400c5996caf.jpg"}, {"type": "image", "uri": "media/robovqa/24a4c99b23454d605e28cab9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0011/robovqa_exec_success", "group_id": "robovqa/11", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 11, "taskvar": "11", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pick up the wrapper", "subtask": "pick up the wrapper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6dec3cca28d0fddf590b3eba.jpg"}, {"type": "image", "uri": "media/robovqa/6dec3cca28d0fddf590b3eba.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0012/robovqa_exec_success", "group_id": "robovqa/12", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 12, "taskvar": "12", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pick up the sanitizer", "subtask": "pick up the sanitizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7db4d5acf3cb41e08f888f52.jpg"}, {"type": "image", "uri": "media/robovqa/49cced3cdfd0158bbbc1fedd.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0013/robovqa_exec_success", "group_id": "robovqa/13", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 13, "taskvar": "13", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "put book on the table", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2459a86ba03df51d694cac79.jpg"}, {"type": "image", "uri": "media/robovqa/597cb96b012c33090ff94f6a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0014/robovqa_exec_success", "group_id": "robovqa/14", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 14, "taskvar": "14", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "move back", "subtask": "move back", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d288b27c64f7fb2be3256765.jpg"}, {"type": "image", "uri": "media/robovqa/cc86ad992d900f491064bfe2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0015/robovqa_exec_success", "group_id": "robovqa/15", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 15, "taskvar": "15", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 226, "role": "outcome"}}
{"state": {"task": "grab the bag", "subtask": "grab the bag", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/9301ad1b0660f1d5e77ad427.jpg"}, {"type": "image", "uri": "media/robovqa/0a86a34a8317c8cc1f42d156.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0016/robovqa_exec_success", "group_id": "robovqa/16", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 16, "taskvar": "16", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "put the cushion properly", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c605a679a961eead2ef8e0a1.jpg"}, {"type": "image", "uri": "media/robovqa/c13c01284c7b1109674e2593.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0017/robovqa_exec_success", "group_id": "robovqa/17", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 17, "taskvar": "17", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 227, "role": "outcome"}}
{"state": {"task": "put the cushion properly", "subtask": "put the cushion properly", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/80060869fe6225298ea20c55.jpg"}, {"type": "image", "uri": "media/robovqa/12b0240026dbd9c9c0b5cd84.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0018/robovqa_exec_success", "group_id": "robovqa/18", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 18, "taskvar": "18", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Pick up the cap", "subtask": "Pick up the cap", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7ddd0d06be66fdc43d2cd0f2.jpg"}, {"type": "image", "uri": "media/robovqa/b60198b185a029c84e432953.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0019/robovqa_exec_success", "group_id": "robovqa/19", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 19, "taskvar": "19", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "move towards the table", "subtask": "move towards the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/173fbc5fa7104978b6f08acf.jpg"}, {"type": "image", "uri": "media/robovqa/8fc2e76971548d8e391f8032.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0020/robovqa_exec_success", "group_id": "robovqa/20", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 20, "taskvar": "20", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "put the cushion properly", "subtask": "grasp cushion", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/08a2d5a67986d13a66464784.jpg"}, {"type": "image", "uri": "media/robovqa/1f9c8c4a5afef9054b18ae0b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0021/robovqa_exec_success", "group_id": "robovqa/21", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 21, "taskvar": "21", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "pick up the pen", "subtask": "pick up the pen", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/0c49ed39d4b5614209f1ef9e.jpg"}, {"type": "image", "uri": "media/robovqa/22972207f4940014ae47673d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0022/robovqa_exec_success", "group_id": "robovqa/22", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 22, "taskvar": "22", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "stack the lid on top of existing stack", "subtask": "stack the lid on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3902abb6cd1f0a582cd5b7e9.jpg"}, {"type": "image", "uri": "media/robovqa/8cfa50001dc9afd1656c9b3c.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0023/robovqa_exec_success", "group_id": "robovqa/23", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 23, "taskvar": "23", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "drop pen into stand", "subtask": "release pen", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f79bf1ab81ee21aef0d84659.jpg"}, {"type": "image", "uri": "media/robovqa/a55a9b7e58c1e91969629024.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0024/robovqa_exec_success", "group_id": "robovqa/24", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 24, "taskvar": "24", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "drop the pen into the pen holder", "subtask": "drop the pen into the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/76b32383a948ec32923090f8.jpg"}, {"type": "image", "uri": "media/robovqa/fe865ab3d226bfe3ea7a722e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0025/robovqa_exec_success", "group_id": "robovqa/25", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 25, "taskvar": "25", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "stack the lid on top of existing stack", "subtask": "stack the lid on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/9e9c9a493b2ecb2a8c5dc7d2.jpg"}, {"type": "image", "uri": "media/robovqa/7036cf4ff1d93bf6933c792f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0026/robovqa_exec_success", "group_id": "robovqa/26", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 26, "taskvar": "26", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "stack the lids on top of existing stack", "subtask": "stack the lids on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7b5c903d4514f562fd447d1e.jpg"}, {"type": "image", "uri": "media/robovqa/ef6c5f019813e6be1d6de9eb.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0027/robovqa_exec_success", "group_id": "robovqa/27", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 27, "taskvar": "27", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "stack the lid on top of existing stack", "subtask": "stack the lid on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/19cb5aef1b646180f8a1bad2.jpg"}, {"type": "image", "uri": "media/robovqa/b06b1c93007051b9d2f19328.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0028/robovqa_exec_success", "group_id": "robovqa/28", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 28, "taskvar": "28", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "drop pen into stand", "subtask": "release pen", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/0f7d07b3fc80cb70a3dde852.jpg"}, {"type": "image", "uri": "media/robovqa/33c1221bc24c3a8078aaf1bf.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0029/robovqa_exec_success", "group_id": "robovqa/29", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 29, "taskvar": "29", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "place the pens into the organizer", "subtask": "place the pens into the organizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/e0ec00b2283a8c7b3d1d0bed.jpg"}, {"type": "image", "uri": "media/robovqa/84330928edebca95078b5b4d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0030/robovqa_exec_success", "group_id": "robovqa/30", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 30, "taskvar": "30", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "move away from the bottom cardboard  box", "subtask": "move away from the bottom cardboard  box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8d16af0aa13b2f5da3fcb547.jpg"}, {"type": "image", "uri": "media/robovqa/fec802e16c3c910f02cd24d9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0031/robovqa_exec_success", "group_id": "robovqa/31", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 31, "taskvar": "31", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "imprecise grasping/pushing", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Go to the cardboard box", "subtask": "Go to the cardboard box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b3d35f00208f05c5f36a4171.jpg"}, {"type": "image", "uri": "media/robovqa/71f27f99da4d40dd044cefd1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0032/robovqa_exec_success", "group_id": "robovqa/32", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 32, "taskvar": "32", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put the pens into the organizer", "subtask": "put the pens into the organizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/13d00beaeae483ab381316c1.jpg"}, {"type": "image", "uri": "media/robovqa/7196b5425a265f8ef332cd0d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0033/robovqa_exec_success", "group_id": "robovqa/33", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 33, "taskvar": "33", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick the brown object from the top cardboard box", "subtask": "pick the brown object from the top cardboard box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/87a9526ea7251a6ecbf9ecbb.jpg"}, {"type": "image", "uri": "media/robovqa/e71abe53624deafa9bc56270.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0034/robovqa_exec_success", "group_id": "robovqa/34", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 34, "taskvar": "34", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "put the pen into the organizer", "subtask": "put the pen into the organizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/965052737afc4c8043b13bad.jpg"}, {"type": "image", "uri": "media/robovqa/9f16d833cf125b97c707125a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0035/robovqa_exec_success", "group_id": "robovqa/35", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 35, "taskvar": "35", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the pads into the organizer", "subtask": "put the pads into the organizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/896da1b9981cba82ddce3657.jpg"}, {"type": "image", "uri": "media/robovqa/7daaec3941107144f88a3d10.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0036/robovqa_exec_success", "group_id": "robovqa/36", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 36, "taskvar": "36", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick the transparent box from the top cardboard box", "subtask": "pick the transparent box from the top cardboard box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/aed91fea2041d98ad897ad85.jpg"}, {"type": "image", "uri": "media/robovqa/b5a95e506dc1b235612f6533.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0037/robovqa_exec_success", "group_id": "robovqa/37", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 37, "taskvar": "37", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "put one apple into the white bowl", "subtask": "put apple in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4647618646c3484e2f48d588.jpg"}, {"type": "image", "uri": "media/robovqa/1997ff87e43bc16d109c88e9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0038/robovqa_exec_success", "group_id": "robovqa/38", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 38, "taskvar": "38", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "close the door", "subtask": "close the door", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/16ccab4150379f497d26bc70.jpg"}, {"type": "image", "uri": "media/robovqa/99b8505d646c08a770964fbf.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0039/robovqa_exec_success", "group_id": "robovqa/39", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 39, "taskvar": "39", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "put pear in the bowl", "subtask": "put pear in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b1ea73bf7d8a6b5a84840d02.jpg"}, {"type": "image", "uri": "media/robovqa/7a23c6480dea6392625b06c9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0040/robovqa_exec_success", "group_id": "robovqa/40", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 40, "taskvar": "40", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "close the door", "subtask": "close the door", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2f46706f0b0b9c503a11ad68.jpg"}, {"type": "image", "uri": "media/robovqa/2f46706f0b0b9c503a11ad68.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0041/robovqa_exec_success", "group_id": "robovqa/41", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 41, "taskvar": "41", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "close the door", "subtask": "close the door", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/505fb80cbb6ebb2968e3a960.jpg"}, {"type": "image", "uri": "media/robovqa/505fb80cbb6ebb2968e3a960.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0042/robovqa_exec_success", "group_id": "robovqa/42", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 42, "taskvar": "42", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "put pear in the bowl", "subtask": "put pear in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/dbd54fe9f3a545b9c7c54fc2.jpg"}, {"type": "image", "uri": "media/robovqa/47faacb9f5d46455b13eabbd.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0043/robovqa_exec_success", "group_id": "robovqa/43", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 43, "taskvar": "43", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put pear in the bowl", "subtask": "put pear in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/84a64d90638b2ff5d162744d.jpg"}, {"type": "image", "uri": "media/robovqa/41e89ef0c336877fe32a35d4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0044/robovqa_exec_success", "group_id": "robovqa/44", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 44, "taskvar": "44", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "close the door", "subtask": "close the door", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/25639ebe517c10dc86126aca.jpg"}, {"type": "image", "uri": "media/robovqa/e827a06406f59af37bcf9d05.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0045/robovqa_exec_success", "group_id": "robovqa/45", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 45, "taskvar": "45", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "Move the cart near to the table", "subtask": "Move the cart near to the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/45e58877fc977bb8b69d8cea.jpg"}, {"type": "image", "uri": "media/robovqa/034e5275e0ece4f192fedf6d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0046/robovqa_exec_success", "group_id": "robovqa/46", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 46, "taskvar": "46", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put packet in drawer", "subtask": "put packet in drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5cc87c118b212b776e63b2f5.jpg"}, {"type": "image", "uri": "media/robovqa/241776eb1a1d21790bb85c80.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0047/robovqa_exec_success", "group_id": "robovqa/47", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 47, "taskvar": "47", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Place the cart in the corner of the room", "subtask": "Place the cart in the corner of the room", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/87257a67dfed2facd07800ae.jpg"}, {"type": "image", "uri": "media/robovqa/14df198b8af2ca8ead535a70.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0048/robovqa_exec_success", "group_id": "robovqa/48", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 48, "taskvar": "48", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "open the drawer of the table", "subtask": "open the drawer of the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/28df9188626874be0d45fbb0.jpg"}, {"type": "image", "uri": "media/robovqa/49f25a30f5fe563e25b142ca.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0049/robovqa_exec_success", "group_id": "robovqa/49", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 49, "taskvar": "49", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the bag in the drawer", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/279150247a9e1cc2d28fe90b.jpg"}, {"type": "image", "uri": "media/robovqa/445384903ade6d103f0ccab9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0050/robovqa_exec_success", "group_id": "robovqa/50", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 50, "taskvar": "50", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "put the white bag in the drawer", "subtask": "put the white bag in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/a3471f25b58f37ebceac0487.jpg"}, {"type": "image", "uri": "media/robovqa/f660fae05195549a35a60704.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0051/robovqa_exec_success", "group_id": "robovqa/51", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 51, "taskvar": "51", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Place the cart in the corner of the room", "subtask": "Place the cart in the corner of the room", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/cfe4f548eefeee5830770f52.jpg"}, {"type": "image", "uri": "media/robovqa/6878d059860a29a23dad92a1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0052/robovqa_exec_success", "group_id": "robovqa/52", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 52, "taskvar": "52", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "Place the book on the top of the existing book", "subtask": "Place the book on the top of the existing book", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/cf1e6bdd9c595229b99afbf4.jpg"}, {"type": "image", "uri": "media/robovqa/a65391d993b05c6ac5892a13.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0053/robovqa_exec_success", "group_id": "robovqa/53", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 53, "taskvar": "53", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 242, "role": "outcome"}}
{"state": {"task": "place the chair away from the table", "subtask": "place the chair away from the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f2b1740e664f1ae291a2094c.jpg"}, {"type": "image", "uri": "media/robovqa/6445ce20b4d3351701fed9f0.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0054/robovqa_exec_success", "group_id": "robovqa/54", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 54, "taskvar": "54", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put monopoly box on the wooden table", "subtask": "put monopoly box on the wooden table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/93b91974949cf15eab092ab5.jpg"}, {"type": "image", "uri": "media/robovqa/d1ee14ee6fc066c93edde46d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0055/robovqa_exec_success", "group_id": "robovqa/55", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 55, "taskvar": "55", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Grasp the white paper", "subtask": "Grasp the white paper", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/addfb91821af3e663734c15c.jpg"}, {"type": "image", "uri": "media/robovqa/45eee46446030e952e54c1b9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0056/robovqa_exec_success", "group_id": "robovqa/56", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 56, "taskvar": "56", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "drag the paper backwards", "subtask": "drag the paper backwards", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/52047f0f384ead2313737b55.jpg"}, {"type": "image", "uri": "media/robovqa/02ee8257cd4b9bfa61d1e3e0.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0057/robovqa_exec_success", "group_id": "robovqa/57", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 57, "taskvar": "57", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "drag the paper backwards", "subtask": "drag the paper backwards", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/ce8e034856a875fd53990b77.jpg"}, {"type": "image", "uri": "media/robovqa/03ee876afaf29c5c173d2e8d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0058/robovqa_exec_success", "group_id": "robovqa/58", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 58, "taskvar": "58", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "put monopoly box on the wooden table", "subtask": "put monopoly box on the wooden table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f53f59ec62f482fe528dd001.jpg"}, {"type": "image", "uri": "media/robovqa/dc0b2e2ce1334278e77f81dc.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0059/robovqa_exec_success", "group_id": "robovqa/59", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 59, "taskvar": "59", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Drag the white paper to the right", "subtask": "Drag the white paper to the right", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d8e267d7744cf88d4f1ce548.jpg"}, {"type": "image", "uri": "media/robovqa/097ca8c0e982f309fa6ed6af.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0060/robovqa_exec_success", "group_id": "robovqa/60", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 60, "taskvar": "60", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "pick the left hoodie arm to fold it", "subtask": "pick the left hoodie arm to fold it", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d61ea5270e4c524c80faf6f6.jpg"}, {"type": "image", "uri": "media/robovqa/40ce37e5d0197e7e5f715528.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0061/robovqa_exec_success", "group_id": "robovqa/61", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 61, "taskvar": "61", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "drop orange in a white bowl", "subtask": "drop orange in a white bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/36c1f4810399d61aace14f0a.jpg"}, {"type": "image", "uri": "media/robovqa/49ff0bb13f1ac689cca8af1f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0062/robovqa_exec_success", "group_id": "robovqa/62", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 62, "taskvar": "62", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put orange in the bowl", "subtask": "put orange in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b0d27a3d3628b5e812271fad.jpg"}, {"type": "image", "uri": "media/robovqa/b0d27a3d3628b5e812271fad.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0063/robovqa_exec_success", "group_id": "robovqa/63", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 63, "taskvar": "63", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "fold the left arm of sweatshirt", "subtask": "fold the left arm of sweatshirt", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/80513e7d6cd5eb11c5f0a874.jpg"}, {"type": "image", "uri": "media/robovqa/72807cfed70459028e432ce7.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0064/robovqa_exec_success", "group_id": "robovqa/64", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 64, "taskvar": "64", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Go towards the left sleeve of sweatshirt", "subtask": "Go towards the left sleeve of sweatshirt", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/180c8e1e69eb14d1ca18a666.jpg"}, {"type": "image", "uri": "media/robovqa/1724aaa9cf77cdb09972b9f6.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0065/robovqa_exec_success", "group_id": "robovqa/65", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 65, "taskvar": "65", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "drop orange in a white bowl", "subtask": "drop orange in a white bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5a8b4737a5b9cf5306591584.jpg"}, {"type": "image", "uri": "media/robovqa/5a8b4737a5b9cf5306591584.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0066/robovqa_exec_success", "group_id": "robovqa/66", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 66, "taskvar": "66", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "fold the hoodie from the down side", "subtask": "fold the hoodie from the down side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/642068eec430662ac7e27ed8.jpg"}, {"type": "image", "uri": "media/robovqa/2a9b9e6fd3a7aad05fe58f68.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0067/robovqa_exec_success", "group_id": "robovqa/67", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 67, "taskvar": "67", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put packet on the table", "subtask": "put packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/1e67195166f5b34604eb90e8.jpg"}, {"type": "image", "uri": "media/robovqa/46526a332b9f6062a131e08d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0068/robovqa_exec_success", "group_id": "robovqa/68", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 68, "taskvar": "68", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "place the red  pens in the pen holder", "subtask": "place the red  pens in the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4f2471fee6cbae0d7a1c8b77.jpg"}, {"type": "image", "uri": "media/robovqa/3337591b7844fbc77ac4c30c.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0069/robovqa_exec_success", "group_id": "robovqa/69", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 69, "taskvar": "69", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "place the black pen in the pen holder", "subtask": "place the black pen in the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/afa61852d7ac46248b70fe29.jpg"}, {"type": "image", "uri": "media/robovqa/6856d9db997cd7df919c96aa.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0070/robovqa_exec_success", "group_id": "robovqa/70", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 70, "taskvar": "70", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put packet on the table", "subtask": "put packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3795d8f755c916edeb3fa3a1.jpg"}, {"type": "image", "uri": "media/robovqa/98abcef7db7a275430b19f90.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0071/robovqa_exec_success", "group_id": "robovqa/71", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 71, "taskvar": "71", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "place the black pen in the pen holder", "subtask": "place the black pen in the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/03118902b2696a94e6ea2f61.jpg"}, {"type": "image", "uri": "media/robovqa/95332eaddaff4b61f06ab0ac.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0072/robovqa_exec_success", "group_id": "robovqa/72", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 72, "taskvar": "72", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put packet on the table", "subtask": "put packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/711d2d23665ad0be2e82cfbe.jpg"}, {"type": "image", "uri": "media/robovqa/f79e5375c8cd41908d51f7b1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0073/robovqa_exec_success", "group_id": "robovqa/73", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 73, "taskvar": "73", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put blue packet in the bowl", "subtask": "put blue packet in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4500744dadccd3c498718407.jpg"}, {"type": "image", "uri": "media/robovqa/da30520b65a29f11463c91ea.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0074/robovqa_exec_success", "group_id": "robovqa/74", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 74, "taskvar": "74", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place can in the tray", "subtask": "place can in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8551a4b11aea2fbd465e4180.jpg"}, {"type": "image", "uri": "media/robovqa/247b0b1a32db7cbc10aa7aa7.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0075/robovqa_exec_success", "group_id": "robovqa/75", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 75, "taskvar": "75", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "place can in the tray", "subtask": "place can in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c4fba228cbdb727f488037b8.jpg"}, {"type": "image", "uri": "media/robovqa/b550a1611013acb540db8dff.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0076/robovqa_exec_success", "group_id": "robovqa/76", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 76, "taskvar": "76", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put blue packet in the bowl", "subtask": "put blue packet in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/1ff2052327f28daea3932234.jpg"}, {"type": "image", "uri": "media/robovqa/62b71053b725dfaf1f32c5e4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0077/robovqa_exec_success", "group_id": "robovqa/77", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 77, "taskvar": "77", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put white sponge in the black stand", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d14bc6fa9a8535b1a9ddfc9a.jpg"}, {"type": "image", "uri": "media/robovqa/ca73c90746c97b9fbb08c94f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0078/robovqa_exec_success", "group_id": "robovqa/78", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 78, "taskvar": "78", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "place pen on the pens", "subtask": "place pen on the pens", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/eae6f3bc5f4bfedbfe51e1af.jpg"}, {"type": "image", "uri": "media/robovqa/508a65697deb8964ea33afef.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0079/robovqa_exec_success", "group_id": "robovqa/79", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 79, "taskvar": "79", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Write something on the white board", "subtask": "Write something on the white board", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/bf5430690f2d0e76d8b1b2f0.jpg"}, {"type": "image", "uri": "media/robovqa/acd699ab57ffd816849e8610.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0080/robovqa_exec_success", "group_id": "robovqa/80", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 80, "taskvar": "80", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place pen on the pens", "subtask": "place pen on the pens", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/bca67d9e2bd590f21fda2742.jpg"}, {"type": "image", "uri": "media/robovqa/bca67d9e2bd590f21fda2742.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0081/robovqa_exec_success", "group_id": "robovqa/81", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 81, "taskvar": "81", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Write f on the white board", "subtask": "Write f on the white board", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/1ddcbfaa3b774a766f736693.jpg"}, {"type": "image", "uri": "media/robovqa/1ddcbfaa3b774a766f736693.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0082/robovqa_exec_success", "group_id": "robovqa/82", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 82, "taskvar": "82", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place pen on the table", "subtask": "place pen on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/730a7c63b40b9af85c608b22.jpg"}, {"type": "image", "uri": "media/robovqa/730a7c63b40b9af85c608b22.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0083/robovqa_exec_success", "group_id": "robovqa/83", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 83, "taskvar": "83", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Write f on the white board", "subtask": "Write f on the white board", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6ccba2796a722903a385a906.jpg"}, {"type": "image", "uri": "media/robovqa/88d975fe45ac7eb7d07875b2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0084/robovqa_exec_success", "group_id": "robovqa/84", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 84, "taskvar": "84", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick up the marker", "subtask": "pick up the marker", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/912505cd1c11b94d1a2222c1.jpg"}, {"type": "image", "uri": "media/robovqa/1fe0e6b524498975c65519ae.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0085/robovqa_exec_success", "group_id": "robovqa/85", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 85, "taskvar": "85", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "place pen on the table", "subtask": "place pen on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b18240756a58f5a09bbf8fed.jpg"}, {"type": "image", "uri": "media/robovqa/d22fc1f5f242350936c120ea.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0086/robovqa_exec_success", "group_id": "robovqa/86", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 86, "taskvar": "86", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "stack the cups on top of existing stack", "subtask": "stack the cups on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fbee8cfa8c2c9719a2befa01.jpg"}, {"type": "image", "uri": "media/robovqa/17833d62fbbb3fe4b7a1f6e1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0087/robovqa_exec_success", "group_id": "robovqa/87", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 87, "taskvar": "87", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "stack the cups on top of existing stack", "subtask": "stack the cups on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f2891b7b16cae0c5ec22f7cb.jpg"}, {"type": "image", "uri": "media/robovqa/f2891b7b16cae0c5ec22f7cb.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0088/robovqa_exec_success", "group_id": "robovqa/88", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 88, "taskvar": "88", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Enter the conference room", "subtask": "open the door of the conference room", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/00b2da267d1a675a5856ffb7.jpg"}, {"type": "image", "uri": "media/robovqa/7f2e971bb77b475f5a7da763.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0089/robovqa_exec_success", "group_id": "robovqa/89", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 89, "taskvar": "89", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "Enter the conference room", "subtask": "open the door of the conference room", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/cb2697eff4c182a5eccf8362.jpg"}, {"type": "image", "uri": "media/robovqa/2460a805864b1693cb92005d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0090/robovqa_exec_success", "group_id": "robovqa/90", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 90, "taskvar": "90", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 233, "role": "outcome"}}
{"state": {"task": "Move the small stack of cups lying on the table on top of an existing stack", "subtask": "stack the cups on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/48ee71b3160fce5291cb9239.jpg"}, {"type": "image", "uri": "media/robovqa/eaee4fdb6a33b7517d6517d2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0091/robovqa_exec_success", "group_id": "robovqa/91", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 91, "taskvar": "91", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 246, "role": "outcome"}}
{"state": {"task": "Pick banana from the platform", "subtask": "Pick banana from the platform", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/73b8f845e5c3b773ffbaeb69.jpg"}, {"type": "image", "uri": "media/robovqa/01b4d44348c82e1880fc36fe.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0092/robovqa_exec_success", "group_id": "robovqa/92", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 92, "taskvar": "92", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "stack the cups on top of existing stack", "subtask": "stack the cups on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8cf5cca04f63c7f5c65b9528.jpg"}, {"type": "image", "uri": "media/robovqa/29e5a14d5d9e2395ba0b31d2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0093/robovqa_exec_success", "group_id": "robovqa/93", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 93, "taskvar": "93", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Pick banana from the platform", "subtask": "Pick banana from the platform", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/dfd5417077f0e38c89c1cbb0.jpg"}, {"type": "image", "uri": "media/robovqa/040c477d5c58bf8f148a74f8.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0094/robovqa_exec_success", "group_id": "robovqa/94", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 94, "taskvar": "94", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "push the trolly near to wall", "subtask": "push the trolly near to wall", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/bfbfe69acc6fd2301773a173.jpg"}, {"type": "image", "uri": "media/robovqa/7f3e62570cf557437b7b29cc.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0095/robovqa_exec_success", "group_id": "robovqa/95", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 95, "taskvar": "95", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "drop the pen into the cart", "subtask": "drop the pen into the cart", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/bf39b894209ab558974ac247.jpg"}, {"type": "image", "uri": "media/robovqa/ffd6ce9116e42c9b35ac1a2a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0096/robovqa_exec_success", "group_id": "robovqa/96", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 96, "taskvar": "96", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the orange in the bowl", "subtask": "put the orange in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/78be058b3696fa66a723e092.jpg"}, {"type": "image", "uri": "media/robovqa/15edcdd9ffa614e36cc25a5b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0097/robovqa_exec_success", "group_id": "robovqa/97", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 97, "taskvar": "97", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the paper on the bowl", "subtask": "place the paper on the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/31e191c8a819d2a2bc2d67b3.jpg"}, {"type": "image", "uri": "media/robovqa/ae262ecd23da612bad4f90a4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0098/robovqa_exec_success", "group_id": "robovqa/98", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 98, "taskvar": "98", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the apple in the bowl to your left", "subtask": "put the apple in the bowl to your left", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/69493686de33e313c677a60c.jpg"}, {"type": "image", "uri": "media/robovqa/6ab2f6c8a0f3177353d30ee1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0099/robovqa_exec_success", "group_id": "robovqa/99", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 99, "taskvar": "99", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "put the pen in trolly", "subtask": "put the pen in trolly", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3ae0795e2a57c4a5414f8014.jpg"}, {"type": "image", "uri": "media/robovqa/ef49d0497913b2266af71152.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0100/robovqa_exec_success", "group_id": "robovqa/100", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 100, "taskvar": "100", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Drop the sticky notes in the cart", "subtask": "Drop the sticky notes in the cart", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2ca509ed239ccd93d8ec0723.jpg"}, {"type": "image", "uri": "media/robovqa/427684fd2146741b1464055a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0101/robovqa_exec_success", "group_id": "robovqa/101", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 101, "taskvar": "101", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the paper on the empty bowl", "subtask": "place the paper on the empty bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/bdf15f53869304b07ff072cb.jpg"}, {"type": "image", "uri": "media/robovqa/36626302495d95201c26b220.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0102/robovqa_exec_success", "group_id": "robovqa/102", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 102, "taskvar": "102", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place  paper clip on pen stand", "subtask": "place  paper clip on pen stand", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/67a1b2fe83747a55f5f4abbf.jpg"}, {"type": "image", "uri": "media/robovqa/64e2210d7b5f495ad1daa567.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0103/robovqa_exec_success", "group_id": "robovqa/103", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 103, "taskvar": "103", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Pick up the glass", "subtask": "Pick up the glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/893df59618c215db5dc6a4e6.jpg"}, {"type": "image", "uri": "media/robovqa/c411acb7b8bad406e5ddd661.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0104/robovqa_exec_success", "group_id": "robovqa/104", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 104, "taskvar": "104", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Place it into the pile of glasses", "subtask": "Place it into the pile of glasses", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/77e5e7c799d8e9ca58efde37.jpg"}, {"type": "image", "uri": "media/robovqa/9fb25353fc28d5616aa99966.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0105/robovqa_exec_success", "group_id": "robovqa/105", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 105, "taskvar": "105", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the orange marker lying on the table into pen holder in the back", "subtask": "place orange marker into pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/e33fb5f29dfe85f3678e10dd.jpg"}, {"type": "image", "uri": "media/robovqa/d2c66874e9ca736e632a40d4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0106/robovqa_exec_success", "group_id": "robovqa/106", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 106, "taskvar": "106", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 242, "role": "outcome"}}
{"state": {"task": "Stack two white cups", "subtask": "release the grasped cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d2d98dcf9e38bfbb7910e186.jpg"}, {"type": "image", "uri": "media/robovqa/0a8de96c4791b10e2ae65cbf.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0107/robovqa_exec_success", "group_id": "robovqa/107", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 107, "taskvar": "107", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put marker in holder", "subtask": "put marker in holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2a73efecbdf5d8ce86a7a80f.jpg"}, {"type": "image", "uri": "media/robovqa/8036777a0063b2503ea10f5b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0108/robovqa_exec_success", "group_id": "robovqa/108", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 108, "taskvar": "108", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "place the stick note into the pen holder", "subtask": "place the stick note into the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5eda13523658f7597598244b.jpg"}, {"type": "image", "uri": "media/robovqa/f458ec6a8a37738921543c17.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0109/robovqa_exec_success", "group_id": "robovqa/109", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 109, "taskvar": "109", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Pick up the bottom right glass", "subtask": "Pick up the bottom right glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/52590e8d474d6d4593cf24aa.jpg"}, {"type": "image", "uri": "media/robovqa/a2e0e18aa59e06be7db6b779.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0110/robovqa_exec_success", "group_id": "robovqa/110", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 110, "taskvar": "110", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "move stapler near another stapler", "subtask": "move stapler near another stapler", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/23909ac5344459161fa6be9f.jpg"}, {"type": "image", "uri": "media/robovqa/b58a70cec1c8cd6556e3fed7.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0111/robovqa_exec_success", "group_id": "robovqa/111", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 111, "taskvar": "111", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the can on the table", "subtask": "move grasped can to table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d605c5a0f155e81b4e50a43c.jpg"}, {"type": "image", "uri": "media/robovqa/89a10900071527c4ece71400.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0112/robovqa_exec_success", "group_id": "robovqa/112", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 112, "taskvar": "112", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the red bull can on the table", "subtask": "place the red bull can on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7d2d503a0a6fabb94b224030.jpg"}, {"type": "image", "uri": "media/robovqa/bf02d5bbbf9df6e142db5b62.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0113/robovqa_exec_success", "group_id": "robovqa/113", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 113, "taskvar": "113", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put tape near pen stand", "subtask": "put tape near pen stand", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d5d91ddf73094fa327b7393c.jpg"}, {"type": "image", "uri": "media/robovqa/37ea9e95fc6d3e04cbf363e6.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0114/robovqa_exec_success", "group_id": "robovqa/114", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 114, "taskvar": "114", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put redbull can on table", "subtask": "grasp redbull can", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/019f4145b8f927b931eb357d.jpg"}, {"type": "image", "uri": "media/robovqa/124fa6c19449c61d6df25f82.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0115/robovqa_exec_success", "group_id": "robovqa/115", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 115, "taskvar": "115", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put redbull can on table", "subtask": "grasp redbull can", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/9601f78b9f298d8892aadbbb.jpg"}, {"type": "image", "uri": "media/robovqa/40b273f1b9180cb32d0b7f55.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0116/robovqa_exec_success", "group_id": "robovqa/116", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 116, "taskvar": "116", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move stapler near another stapler", "subtask": "move stapler near another stapler", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3aebb43071be5c1149bde12f.jpg"}, {"type": "image", "uri": "media/robovqa/590e9a2ec241efabfbf867d9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0117/robovqa_exec_success", "group_id": "robovqa/117", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 117, "taskvar": "117", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put tape near pen stand", "subtask": "put tape near pen stand", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/cbcb5f61f0147511b1ddec90.jpg"}, {"type": "image", "uri": "media/robovqa/4247ebc39731983aaca41371.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0118/robovqa_exec_success", "group_id": "robovqa/118", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 118, "taskvar": "118", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Place the paper cup upside down on the table", "subtask": "Place the paper cup upside down on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f0df1caf46ad05dda875d345.jpg"}, {"type": "image", "uri": "media/robovqa/9507de65d4a668b67df6e933.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0119/robovqa_exec_success", "group_id": "robovqa/119", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 119, "taskvar": "119", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "put the wipes on the table", "subtask": "put the wipes on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/da9aa8963ddd3f373debb8f6.jpg"}, {"type": "image", "uri": "media/robovqa/9d6ebf271a6959a450cd6cb5.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0120/robovqa_exec_success", "group_id": "robovqa/120", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 120, "taskvar": "120", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "clean the tab", "subtask": "clean the tab", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b7c6994f50da03b0dfabde0a.jpg"}, {"type": "image", "uri": "media/robovqa/cf10536bb56a7280904e1cad.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0121/robovqa_exec_success", "group_id": "robovqa/121", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 121, "taskvar": "121", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "Put the orange fruit on the desk", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/50ea79e5c47178812dea36f2.jpg"}, {"type": "image", "uri": "media/robovqa/7c3679c531925a106066901f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0122/robovqa_exec_success", "group_id": "robovqa/122", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 122, "taskvar": "122", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Put the grapes in the paper cup", "subtask": "Put the grapes in the paper cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7061d29fbe58073d106f598a.jpg"}, {"type": "image", "uri": "media/robovqa/f78745e54e6f15f041a1f805.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0123/robovqa_exec_success", "group_id": "robovqa/123", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 123, "taskvar": "123", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place wipes on the table", "subtask": "place wipes on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2ecd59f0c7959e2e80c29484.jpg"}, {"type": "image", "uri": "media/robovqa/e3b1ff567bb8c97f997c50e1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0124/robovqa_exec_success", "group_id": "robovqa/124", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 124, "taskvar": "124", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put the packet on the table", "subtask": "put the packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/ee94f14aed62cbae4904cf1e.jpg"}, {"type": "image", "uri": "media/robovqa/ee94f14aed62cbae4904cf1e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0125/robovqa_exec_success", "group_id": "robovqa/125", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 125, "taskvar": "125", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Place the paper cup on the table", "subtask": "Place the paper cup on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/9d6a5876f577bbb98dc18fd3.jpg"}, {"type": "image", "uri": "media/robovqa/15674b8372d49346074af8d3.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0126/robovqa_exec_success", "group_id": "robovqa/126", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 126, "taskvar": "126", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "drop the green packet into the organizer", "subtask": "drop the green packet into the organizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c01a6c9b65e17b7b0b91f8c7.jpg"}, {"type": "image", "uri": "media/robovqa/020df92e8bb4b9038eda5991.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0127/robovqa_exec_success", "group_id": "robovqa/127", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 127, "taskvar": "127", "episode_id": null, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the apple on the table", "subtask": "Put the apple on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/40f0972103b6401734409b38.jpg"}, {"type": "image", "uri": "media/robovqa/9414094df92b02963c0c76cc.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0128/robovqa_exec_success", "group_id": "robovqa/128", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 128, "taskvar": "128", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the two brown sugar packets lying on the table into the organiser", "subtask": "Put  brown sugar packet in the organiser", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d3e5b807ea5112ae9c9b08f2.jpg"}, {"type": "image", "uri": "media/robovqa/006fffce88f301f35849d0e7.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0129/robovqa_exec_success", "group_id": "robovqa/129", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 129, "taskvar": "129", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 243, "role": "outcome"}}
{"state": {"task": "drop the white packet into the organizer", "subtask": "drop the white packet into the organizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/97602a008e69c8c843be6fdf.jpg"}, {"type": "image", "uri": "media/robovqa/111fa43bb01657b17d565501.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0130/robovqa_exec_success", "group_id": "robovqa/130", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 130, "taskvar": "130", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Place one of the apple on the table", "subtask": "Place one of the apple on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4abd44a3f1196bf5798390f1.jpg"}, {"type": "image", "uri": "media/robovqa/47c66e266d6e90f61c7617d7.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0131/robovqa_exec_success", "group_id": "robovqa/131", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 131, "taskvar": "131", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put the white packet into the organizer", "subtask": "put the white packet into the organizer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c8c9877e539f7583656adcb0.jpg"}, {"type": "image", "uri": "media/robovqa/45eb335ba75a7262dd4d5043.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0132/robovqa_exec_success", "group_id": "robovqa/132", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 132, "taskvar": "132", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Empty the bowl of apple", "subtask": "Empty the bowl of apple", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fbbe2d86f4afef071c5de612.jpg"}, {"type": "image", "uri": "media/robovqa/b0b03a8480516d915b78e334.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0133/robovqa_exec_success", "group_id": "robovqa/133", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 133, "taskvar": "133", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Put the apple on the table", "subtask": "Put the apple on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/737e7b50e384617b394f7eeb.jpg"}, {"type": "image", "uri": "media/robovqa/a8190e9de03276c0b6818200.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0134/robovqa_exec_success", "group_id": "robovqa/134", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 134, "taskvar": "134", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/45503eba2e500138a65b8a6c.jpg"}, {"type": "image", "uri": "media/robovqa/85afd475b9f72af61945fad2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0135/robovqa_exec_success", "group_id": "robovqa/135", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 135, "taskvar": "135", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "put red plate in the drawer", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fd93b71e781ffe537f20c15d.jpg"}, {"type": "image", "uri": "media/robovqa/e316e2f00277674debf1e631.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0136/robovqa_exec_success", "group_id": "robovqa/136", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 136, "taskvar": "136", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the red book on the table", "subtask": "place the book on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/50f523776029b3268c2094c8.jpg"}, {"type": "image", "uri": "media/robovqa/03dca9aec7699e1127b4b207.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0137/robovqa_exec_success", "group_id": "robovqa/137", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 137, "taskvar": "137", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "place the red book on the table", "subtask": "place the book on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7e88ba56abab4dcd2c5b7cea.jpg"}, {"type": "image", "uri": "media/robovqa/7e88ba56abab4dcd2c5b7cea.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0138/robovqa_exec_success", "group_id": "robovqa/138", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 138, "taskvar": "138", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "put plate in the drawer", "subtask": "put plate in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/ba207b08642936ad24b0f3bb.jpg"}, {"type": "image", "uri": "media/robovqa/c533df8548415f5445ace973.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0139/robovqa_exec_success", "group_id": "robovqa/139", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 139, "taskvar": "139", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "place the book on the table", "subtask": "place the book on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/36200e22e7a85ff4c2c1735e.jpg"}, {"type": "image", "uri": "media/robovqa/977a084860d7e64fe44a259f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0140/robovqa_exec_success", "group_id": "robovqa/140", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 140, "taskvar": "140", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put plate in the drawer", "subtask": "put plate in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f1cfde9cee6db442bde432fa.jpg"}, {"type": "image", "uri": "media/robovqa/01cdcaec81ddbb066ed9493e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0141/robovqa_exec_success", "group_id": "robovqa/141", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 141, "taskvar": "141", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "place the red book on the table", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/aa5025061eb794afea13581d.jpg"}, {"type": "image", "uri": "media/robovqa/d3f4d382aa3b0a1b340f1ecd.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0142/robovqa_exec_success", "group_id": "robovqa/142", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 142, "taskvar": "142", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "put the packet in the bowl", "subtask": "move the grasped packet in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3419889cd766972cf245969d.jpg"}, {"type": "image", "uri": "media/robovqa/1f7bda5a6e2ce6f8b988fd5e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0143/robovqa_exec_success", "group_id": "robovqa/143", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 143, "taskvar": "143", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "label_conflict": "reward 1 but Guardian category 'no_progress'; label follows the reward (= Guardian's gold True/False answer)", "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the apple in the container", "subtask": "put the apple in the container", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4d8bf9fe1184c9b1672ca6d9.jpg"}, {"type": "image", "uri": "media/robovqa/4d8bf9fe1184c9b1672ca6d9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0144/robovqa_exec_success", "group_id": "robovqa/144", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 144, "taskvar": "144", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place orange in large bowl", "subtask": "place orange in large bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/367a6376ee06d3f8621a0264.jpg"}, {"type": "image", "uri": "media/robovqa/c8941e4fffe3033bb11317dc.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0145/robovqa_exec_success", "group_id": "robovqa/145", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 145, "taskvar": "145", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Place the fruit in the tray", "subtask": "Place the fruit in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/ae7c21ba9a9740ef92d7969b.jpg"}, {"type": "image", "uri": "media/robovqa/a842d1aad8ef0015455e475f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0146/robovqa_exec_success", "group_id": "robovqa/146", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 146, "taskvar": "146", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the apple into the basket", "subtask": "place the apple into the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/048d0274537fda10c440d148.jpg"}, {"type": "image", "uri": "media/robovqa/c9d213726a3df213a63f4a4b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0147/robovqa_exec_success", "group_id": "robovqa/147", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 147, "taskvar": "147", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the apple in the gray container", "subtask": "move grasped apple to gray container", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d2284b67ca3a7031cdad3d70.jpg"}, {"type": "image", "uri": "media/robovqa/ccfcd300185fac2ec22a1294.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0148/robovqa_exec_success", "group_id": "robovqa/148", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 148, "taskvar": "148", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the orange in the bowl", "subtask": "put the orange in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/461f77fbefea5e06d1e475dc.jpg"}, {"type": "image", "uri": "media/robovqa/ec1fa71dd5738f0b001c42d1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0149/robovqa_exec_success", "group_id": "robovqa/149", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 149, "taskvar": "149", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the orange in the bowl", "subtask": "put the orange in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/42fbc5a6e1bf3c0252e41e1a.jpg"}, {"type": "image", "uri": "media/robovqa/5739db06726d4b220036d9ee.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0150/robovqa_exec_success", "group_id": "robovqa/150", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 150, "taskvar": "150", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the wet tissue on the table", "subtask": "place the wet tissue on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8a47122c59ca6ae22cbdb4da.jpg"}, {"type": "image", "uri": "media/robovqa/8c107bd8d56538abc33cd0f5.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0151/robovqa_exec_success", "group_id": "robovqa/151", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 151, "taskvar": "151", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "move the chair forward", "subtask": "grasp the chair", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/54b6ccdf68d9c1e0e39b2247.jpg"}, {"type": "image", "uri": "media/robovqa/a05e105fe7dc3e16bb7eb32e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0152/robovqa_exec_success", "group_id": "robovqa/152", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 152, "taskvar": "152", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "move the chair forward", "subtask": "move the chair forward", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/0ffbcaae54a07a64f183e95c.jpg"}, {"type": "image", "uri": "media/robovqa/c1e99bcc727090b2fe68c89b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0153/robovqa_exec_success", "group_id": "robovqa/153", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 153, "taskvar": "153", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Put the lids on the table", "subtask": "Put the lids on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/611903feee420f307be69e05.jpg"}, {"type": "image", "uri": "media/robovqa/459991a6cda8acd76564b97a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0154/robovqa_exec_success", "group_id": "robovqa/154", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 154, "taskvar": "154", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the wet tissue on the table", "subtask": "place the wet tissue on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6a699ede394e3a3bbe9ed304.jpg"}, {"type": "image", "uri": "media/robovqa/04da02c86a332bd5c17967a2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0155/robovqa_exec_success", "group_id": "robovqa/155", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 155, "taskvar": "155", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the lids on the table", "subtask": "grasp one lid", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/945fb96237ae6bca12724c5c.jpg"}, {"type": "image", "uri": "media/robovqa/b12be2a9da6b24e962ccf0ec.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0156/robovqa_exec_success", "group_id": "robovqa/156", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 156, "taskvar": "156", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Put the lids on the table", "subtask": "Put the lids on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/e8915df1f7e9d3b78f923ba3.jpg"}, {"type": "image", "uri": "media/robovqa/8749dfb70746669d2e20d8f0.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0157/robovqa_exec_success", "group_id": "robovqa/157", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 157, "taskvar": "157", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the pear on the table", "subtask": "place the pear on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c499e1b2b8aaf168f93014bf.jpg"}, {"type": "image", "uri": "media/robovqa/0a2d6ad5d682b085412c5751.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0158/robovqa_exec_success", "group_id": "robovqa/158", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 158, "taskvar": "158", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place one of the red apple on the table", "subtask": "place one of the red apple on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f3776e76971e2dbdc19eaf8d.jpg"}, {"type": "image", "uri": "media/robovqa/bddacc358fca068df156ff21.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0159/robovqa_exec_success", "group_id": "robovqa/159", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 159, "taskvar": "159", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "Stack the glasses on top of existing stack", "subtask": "Stack the glasses on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/942e6a6763b742d6610918a1.jpg"}, {"type": "image", "uri": "media/robovqa/443cf1ded47a77a622520e64.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0160/robovqa_exec_success", "group_id": "robovqa/160", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 160, "taskvar": "160", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Stack the lid on top of existing stack", "subtask": "Stack the lid on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4dacc81cc396c8bca65a2379.jpg"}, {"type": "image", "uri": "media/robovqa/6b979bc066095bb309b1a08d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0161/robovqa_exec_success", "group_id": "robovqa/161", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 161, "taskvar": "161", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Stack the lid on top of existing stack", "subtask": "Stack the lid on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/89a0a1e2e8f8eb04afa7e86d.jpg"}, {"type": "image", "uri": "media/robovqa/75ebaa6ee61d756c7e32542f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0162/robovqa_exec_success", "group_id": "robovqa/162", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 162, "taskvar": "162", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Stack the lid on top of existing stack", "subtask": "Stack the lid on top of existing stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/e7dfef9de6c350585f171bd8.jpg"}, {"type": "image", "uri": "media/robovqa/1732889fbbf942ad6bfe5262.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0163/robovqa_exec_success", "group_id": "robovqa/163", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 163, "taskvar": "163", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "place the packet on the table", "subtask": "place the packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c1973bee3795ea4f709bc12a.jpg"}, {"type": "image", "uri": "media/robovqa/17e202cb6446c7b16e871b9d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0164/robovqa_exec_success", "group_id": "robovqa/164", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 164, "taskvar": "164", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the packets on the table", "subtask": "Put the packets on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/072bd5dd1f5939f6f799c21f.jpg"}, {"type": "image", "uri": "media/robovqa/c9782c3091af7cc1ab3317f0.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0165/robovqa_exec_success", "group_id": "robovqa/165", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 165, "taskvar": "165", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the book in the shelf", "subtask": "Put the book in the shelf", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d4eefe91735efd9911f8a202.jpg"}, {"type": "image", "uri": "media/robovqa/969b2ea0de9585a8cd74adcd.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0166/robovqa_exec_success", "group_id": "robovqa/166", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 166, "taskvar": "166", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the book in the shelf", "subtask": "Put the book in the shelf", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/ad997989897dcdfaa71db7fc.jpg"}, {"type": "image", "uri": "media/robovqa/6756e79f50e7bb1fe9ec6fec.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0167/robovqa_exec_success", "group_id": "robovqa/167", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 167, "taskvar": "167", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Move the chair towards the table", "subtask": "grasp chair", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d502963e7a64e31d0d5fbf78.jpg"}, {"type": "image", "uri": "media/robovqa/7dab669ac2b8c533faf4d47f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0168/robovqa_exec_success", "group_id": "robovqa/168", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 168, "taskvar": "168", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "Put the book in the shelf", "subtask": "Put the book in the shelf", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/09014768e88402244b663866.jpg"}, {"type": "image", "uri": "media/robovqa/14d32ad14e2d9c90d7cc6276.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0169/robovqa_exec_success", "group_id": "robovqa/169", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 169, "taskvar": "169", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "empty onto the table the gray container from its potato packets", "subtask": "empty onto the table the gray container from its potato packets", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2faaa39ad225e61e88e6cf10.jpg"}, {"type": "image", "uri": "media/robovqa/a3bc86f89af38c848e9b0096.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0170/robovqa_exec_success", "group_id": "robovqa/170", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 170, "taskvar": "170", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 244, "role": "outcome"}}
{"state": {"task": "place the packets on the table", "subtask": "place the packets on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b330701a592e58a3d1326f1c.jpg"}, {"type": "image", "uri": "media/robovqa/b330701a592e58a3d1326f1c.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0171/robovqa_exec_success", "group_id": "robovqa/171", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 171, "taskvar": "171", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the block in the box", "subtask": "put the block in the box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5092e5f89f9eeee3f1e75bb6.jpg"}, {"type": "image", "uri": "media/robovqa/43112783fe664e81b9da328a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0172/robovqa_exec_success", "group_id": "robovqa/172", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 172, "taskvar": "172", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick up the tissues", "subtask": "pick up the tissues", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/854aa30c3022f31d6c28b120.jpg"}, {"type": "image", "uri": "media/robovqa/6e9297b8903f89daf906d2bb.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0173/robovqa_exec_success", "group_id": "robovqa/173", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 173, "taskvar": "173", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Take tissue", "subtask": "Take tissue", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/92899cc35db2f39572c9ccd0.jpg"}, {"type": "image", "uri": "media/robovqa/92899cc35db2f39572c9ccd0.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0174/robovqa_exec_success", "group_id": "robovqa/174", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 174, "taskvar": "174", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 226, "role": "outcome"}}
{"state": {"task": "put the spoon in the white cup", "subtask": "put the spoon in the white cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/376ce01adf93faa6944c35e0.jpg"}, {"type": "image", "uri": "media/robovqa/9a29d23692d47b441e0326dd.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0175/robovqa_exec_success", "group_id": "robovqa/175", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 175, "taskvar": "175", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Insert the spoon into the chobani box", "subtask": "Insert the spoon into the chobani box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/90cf8fbc142a0328ae5f768b.jpg"}, {"type": "image", "uri": "media/robovqa/e2184892c1952ccc12cfab22.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0176/robovqa_exec_success", "group_id": "robovqa/176", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 176, "taskvar": "176", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "stack the plastic glass", "subtask": "stack the plastic glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4e49dc855d0b133341c3bbe8.jpg"}, {"type": "image", "uri": "media/robovqa/442b5b993953a43e17dfff92.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0177/robovqa_exec_success", "group_id": "robovqa/177", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 177, "taskvar": "177", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "place block in the box", "subtask": "grasp block", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/96e0d4955975f564013c98df.jpg"}, {"type": "image", "uri": "media/robovqa/9366399f083ec9f92a8d0f75.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0178/robovqa_exec_success", "group_id": "robovqa/178", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 178, "taskvar": "178", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Put the yellow color object in the box", "subtask": "Put the yellow color object in the box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5c6868745d77bf27566dee49.jpg"}, {"type": "image", "uri": "media/robovqa/b45685b3c1fbba7934823524.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0179/robovqa_exec_success", "group_id": "robovqa/179", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 179, "taskvar": "179", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put packet on the packets", "subtask": "put packet on the packets", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d9a25c8c829a19707499fe4c.jpg"}, {"type": "image", "uri": "media/robovqa/d9a25c8c829a19707499fe4c.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0180/robovqa_exec_success", "group_id": "robovqa/180", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 180, "taskvar": "180", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put the plastic fork into the black container", "subtask": "move grasped plastic fork to black container", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/16b54d15e7b4e1f4cc8030b6.jpg"}, {"type": "image", "uri": "media/robovqa/6f86951c6f9a40d2d11459b2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0181/robovqa_exec_success", "group_id": "robovqa/181", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 181, "taskvar": "181", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put the fork spoon in the pen holder", "subtask": "put the fork spoon in the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/18806ac13c7bef181fd901f2.jpg"}, {"type": "image", "uri": "media/robovqa/1d03aad9aa62c907b6fd00ac.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0182/robovqa_exec_success", "group_id": "robovqa/182", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 182, "taskvar": "182", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put packet on the packets", "subtask": "put packet on the packets", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fd0f694cbdb04074607413dd.jpg"}, {"type": "image", "uri": "media/robovqa/6bd6e9b578717ea0953069a1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0183/robovqa_exec_success", "group_id": "robovqa/183", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 183, "taskvar": "183", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put packets from black basket to the table", "subtask": "put packets from black basket to the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d7548d1fa71268188589eced.jpg"}, {"type": "image", "uri": "media/robovqa/e70e50555096ed38e6a1ea8d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0184/robovqa_exec_success", "group_id": "robovqa/184", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 184, "taskvar": "184", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put the plastic fork into the pen holder", "subtask": "put the plastic fork into the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/47dfec98a3676bb18b7b2c1b.jpg"}, {"type": "image", "uri": "media/robovqa/3e5a0d2649456a5d8b73303f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0185/robovqa_exec_success", "group_id": "robovqa/185", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 185, "taskvar": "185", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put the plastic fork into the pen holder", "subtask": "put the plastic fork into the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7371c22ad94618846d0e3390.jpg"}, {"type": "image", "uri": "media/robovqa/7371c22ad94618846d0e3390.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0186/robovqa_exec_success", "group_id": "robovqa/186", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 186, "taskvar": "186", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put packet on the packets", "subtask": "put packet on the packets", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c5c6d27662a5fcb605545372.jpg"}, {"type": "image", "uri": "media/robovqa/2171acc2dafca6906a2d2f04.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0187/robovqa_exec_success", "group_id": "robovqa/187", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 187, "taskvar": "187", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put the orange in the bowl", "subtask": "put the orange in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5d3a0e440041c5af71cec5ae.jpg"}, {"type": "image", "uri": "media/robovqa/cf04cfc2c33204bf590ad7c5.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0188/robovqa_exec_success", "group_id": "robovqa/188", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 188, "taskvar": "188", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the egg in the white front bowl", "subtask": "move the grasped egg in the white front bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/9ff4eea61aa6144410830e14.jpg"}, {"type": "image", "uri": "media/robovqa/da8586dc322fa24ed61ed4ef.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0189/robovqa_exec_success", "group_id": "robovqa/189", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 189, "taskvar": "189", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "clean the table with the tissue", "subtask": "grasp tissue", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/aac2fe4eeafa51e7bdd776a1.jpg"}, {"type": "image", "uri": "media/robovqa/a7e809d07dd55f7731176f5e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0190/robovqa_exec_success", "group_id": "robovqa/190", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 190, "taskvar": "190", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "put the white bowl out of the stack", "subtask": "put the white bowl out of the stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/70c9ada33a73342dfac9f422.jpg"}, {"type": "image", "uri": "media/robovqa/2653b846999afb4555203759.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0191/robovqa_exec_success", "group_id": "robovqa/191", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 191, "taskvar": "191", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put the white tissue on the table", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2be9c38fcb7d40fc23361682.jpg"}, {"type": "image", "uri": "media/robovqa/409b2f9812299679989ff699.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0192/robovqa_exec_success", "group_id": "robovqa/192", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 192, "taskvar": "192", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "put orange in the bowl", "subtask": "grasp orange", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b8cd1f80450d68c81ae0933c.jpg"}, {"type": "image", "uri": "media/robovqa/9699f141fa639e95246caae2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0193/robovqa_exec_success", "group_id": "robovqa/193", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 193, "taskvar": "193", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Move towards the black cable", "subtask": "Move towards the black cable", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/896405ecca798be15c3df12e.jpg"}, {"type": "image", "uri": "media/robovqa/a5fcd86540a40de3da0e9a55.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0194/robovqa_exec_success", "group_id": "robovqa/194", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 194, "taskvar": "194", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Close the trap door", "subtask": "grasp door handle", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5df32c5c17af29c79aacfcdf.jpg"}, {"type": "image", "uri": "media/robovqa/64fa0a89002d8c560b23c8d6.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0195/robovqa_exec_success", "group_id": "robovqa/195", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 195, "taskvar": "195", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "place the tray in the shelf", "subtask": "release tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/0caf4d0d3e47e7d04a8840d7.jpg"}, {"type": "image", "uri": "media/robovqa/cc19fc74fd7272c7f0d0b39b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0196/robovqa_exec_success", "group_id": "robovqa/196", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 196, "taskvar": "196", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "place the tray in the shelf", "subtask": "place the tray in the shelf", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/107d924fc3e2015e9ef61f8d.jpg"}, {"type": "image", "uri": "media/robovqa/895bc9dbf1228753fc69bc53.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0197/robovqa_exec_success", "group_id": "robovqa/197", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 197, "taskvar": "197", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the tray in the shelf", "subtask": "place the tray in the shelf", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f9e29f7570a1395da8a2915a.jpg"}, {"type": "image", "uri": "media/robovqa/32cd0c431e8ac4dfb36078ab.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0198/robovqa_exec_success", "group_id": "robovqa/198", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 198, "taskvar": "198", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Slightly move the brown cable", "subtask": "Slightly move the brown cable", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fc3a2fc8338187fb7f774b80.jpg"}, {"type": "image", "uri": "media/robovqa/834416e538adeb1c74c5acb1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0199/robovqa_exec_success", "group_id": "robovqa/199", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 199, "taskvar": "199", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the tray in the shelf", "subtask": "place the tray in the shelf", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7feb310d938db58470cd3b19.jpg"}, {"type": "image", "uri": "media/robovqa/37798847dca2bb4ecbfbbd1b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0200/robovqa_exec_success", "group_id": "robovqa/200", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 200, "taskvar": "200", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the red cable into the trap door", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/a7881c1b65b933fd91a83a87.jpg"}, {"type": "image", "uri": "media/robovqa/70ce286c09f537308168ec16.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0201/robovqa_exec_success", "group_id": "robovqa/201", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 201, "taskvar": "201", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "pull the drawer", "subtask": "pull the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f18d9cc5b1f37ad8386a6406.jpg"}, {"type": "image", "uri": "media/robovqa/4c91fba8bd4af29a96599e18.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0202/robovqa_exec_success", "group_id": "robovqa/202", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 202, "taskvar": "202", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "pick up the can from the tray", "subtask": "pick up the can from the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/67107a6460afe6678fcf624a.jpg"}, {"type": "image", "uri": "media/robovqa/c643daf7ca87235d09b9380c.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0203/robovqa_exec_success", "group_id": "robovqa/203", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 203, "taskvar": "203", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the pen holder at the sticky notes", "subtask": "Put the pen holder at the sticky notes", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/748c5203e2d51a6bacbba90c.jpg"}, {"type": "image", "uri": "media/robovqa/6f3a74e2baf28d2d2fc3a9a7.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0204/robovqa_exec_success", "group_id": "robovqa/204", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 204, "taskvar": "204", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "put the blue marker in the drawer", "subtask": "put the blue marker in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2702f203f356a52e17b0d620.jpg"}, {"type": "image", "uri": "media/robovqa/3b0156097910b7de5afddecf.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0205/robovqa_exec_success", "group_id": "robovqa/205", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 205, "taskvar": "205", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the red pen in the drawer", "subtask": "put the red pen in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/e5171471bba5c8042cd631ec.jpg"}, {"type": "image", "uri": "media/robovqa/88c5db011813505bfb9a4ce6.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0206/robovqa_exec_success", "group_id": "robovqa/206", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 206, "taskvar": "206", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place onto the table the remaining can that is in the container", "subtask": "grasp can", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/ec17df0db4330f0eee4049a5.jpg"}, {"type": "image", "uri": "media/robovqa/48f0fc8cc1681f7f6b37e23c.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0207/robovqa_exec_success", "group_id": "robovqa/207", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 207, "taskvar": "207", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "place that can on the table", "subtask": "place that can on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/eeed57c9c417e046c4e01020.jpg"}, {"type": "image", "uri": "media/robovqa/92ed651e7fb7787a0162202a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0208/robovqa_exec_success", "group_id": "robovqa/208", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 208, "taskvar": "208", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the fork into the holder", "subtask": "place the fork into the holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/950d985665ea1c382d5e8edf.jpg"}, {"type": "image", "uri": "media/robovqa/3307208afc8f3aa9b954969f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0209/robovqa_exec_success", "group_id": "robovqa/209", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 209, "taskvar": "209", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the fork in the holder", "subtask": "Put the fork in the holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5e5ed8ca3b6cd01e54c50ffa.jpg"}, {"type": "image", "uri": "media/robovqa/c683010634df2fcc7f0e6b97.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0210/robovqa_exec_success", "group_id": "robovqa/210", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 210, "taskvar": "210", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the cherries in the bowl", "subtask": "place the cherries in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/aa75ee45eb7dff02ff44e23b.jpg"}, {"type": "image", "uri": "media/robovqa/7e80efab6306e73e89f18db3.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0211/robovqa_exec_success", "group_id": "robovqa/211", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 211, "taskvar": "211", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the black packet into the tray", "subtask": "place the black packet into the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/05b9d5f3723b0799bea87730.jpg"}, {"type": "image", "uri": "media/robovqa/7fe377b45197fceb11929402.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0212/robovqa_exec_success", "group_id": "robovqa/212", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 212, "taskvar": "212", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the fork in the holder", "subtask": "Put the fork in the holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/cc63888082919aa9ed735c34.jpg"}, {"type": "image", "uri": "media/robovqa/6fe872c5ceada77d2a4a0d57.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0213/robovqa_exec_success", "group_id": "robovqa/213", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 213, "taskvar": "213", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the green apple from the left of the table into the bowl", "subtask": "place the green apple from the left of the table into the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/9922d2b72ca7d19506ba1ec3.jpg"}, {"type": "image", "uri": "media/robovqa/9922d2b72ca7d19506ba1ec3.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0214/robovqa_exec_success", "group_id": "robovqa/214", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 214, "taskvar": "214", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 248, "role": "outcome"}}
{"state": {"task": "place the apple in the bowl", "subtask": "place the apple in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/119f40b15dd7af990b58abd9.jpg"}, {"type": "image", "uri": "media/robovqa/5bd8d9dd719aac95b5a62fc1.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0215/robovqa_exec_success", "group_id": "robovqa/215", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 215, "taskvar": "215", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the grapes in the bowl", "subtask": "place the grapes in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d0f5e582d48b89170ee69569.jpg"}, {"type": "image", "uri": "media/robovqa/315577ac5987c507c2c8637e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0216/robovqa_exec_success", "group_id": "robovqa/216", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 216, "taskvar": "216", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick the cap", "subtask": "pick the cap", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5d752ffa09d8e94dfc5b4b48.jpg"}, {"type": "image", "uri": "media/robovqa/b9ab571069fa5071274b2f61.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0217/robovqa_exec_success", "group_id": "robovqa/217", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 217, "taskvar": "217", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "pick the cap", "subtask": "pick the cap", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f099fa19f42dbe343447993d.jpg"}, {"type": "image", "uri": "media/robovqa/cf83e2deb81a7c60df6997e9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0218/robovqa_exec_success", "group_id": "robovqa/218", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 218, "taskvar": "218", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "pick the cap", "subtask": "pick the cap", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6a6f4c9e9c85bac61e88ec25.jpg"}, {"type": "image", "uri": "media/robovqa/6402af8016c086c13ef4ac4a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0219/robovqa_exec_success", "group_id": "robovqa/219", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 219, "taskvar": "219", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "hang the bag on the stand", "subtask": "hang the bag on the stand", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/61a1a0d931829bddec9ffa92.jpg"}, {"type": "image", "uri": "media/robovqa/e39cedfd99564369890793b5.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0220/robovqa_exec_success", "group_id": "robovqa/220", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 220, "taskvar": "220", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place one red pen on the table", "subtask": "place one red pen on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/708113d3ecab5ba9e1e2155d.jpg"}, {"type": "image", "uri": "media/robovqa/eb6bf055106779baad972207.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0221/robovqa_exec_success", "group_id": "robovqa/221", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 221, "taskvar": "221", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "open the shelf door", "subtask": "open the shelf door", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/9741bd17b75d253cdd146f73.jpg"}, {"type": "image", "uri": "media/robovqa/ee0d7f059474437898722372.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0222/robovqa_exec_success", "group_id": "robovqa/222", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 222, "taskvar": "222", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "place green marker in the rack", "subtask": "place green marker in the rack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/bcc236fb26a8404b9aeaf8c6.jpg"}, {"type": "image", "uri": "media/robovqa/c27a88d45ca84ae0a4bbf797.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0223/robovqa_exec_success", "group_id": "robovqa/223", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 223, "taskvar": "223", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "close the pink marker drawer", "subtask": "close the pink marker drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/29c0b7a88304a0362e99b4a2.jpg"}, {"type": "image", "uri": "media/robovqa/77b11afbe13485427eb2fbbc.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0224/robovqa_exec_success", "group_id": "robovqa/224", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 224, "taskvar": "224", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "pull the drawer", "subtask": "pull the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3bef1c4a27ea44bad276c87a.jpg"}, {"type": "image", "uri": "media/robovqa/43d8558dfb936ec772c57e3d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0225/robovqa_exec_success", "group_id": "robovqa/225", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 225, "taskvar": "225", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "adjust the pen", "subtask": "adjust the pen", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/18d6be10015a975b2c77de67.jpg"}, {"type": "image", "uri": "media/robovqa/9d3d180bd4aeaf3ed613abb9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0226/robovqa_exec_success", "group_id": "robovqa/226", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 226, "taskvar": "226", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b179b2b33d8f3d43b6ed461a.jpg"}, {"type": "image", "uri": "media/robovqa/068c718045df4254df3aadc9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0227/robovqa_exec_success", "group_id": "robovqa/227", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 227, "taskvar": "227", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "move the right white tray next to the left white packet", "subtask": "move the right white tray next to the left white packet", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/cf3cf4a8257760f2221c47da.jpg"}, {"type": "image", "uri": "media/robovqa/4d509c51ed8cc5e13a99d647.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0228/robovqa_exec_success", "group_id": "robovqa/228", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 228, "taskvar": "228", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 244, "role": "outcome"}}
{"state": {"task": "place pen in the tray", "subtask": "place pen in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8638701df8682e4e839cf8d2.jpg"}, {"type": "image", "uri": "media/robovqa/dfd863543be280b1b66efa06.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0229/robovqa_exec_success", "group_id": "robovqa/229", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 229, "taskvar": "229", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "lock the drawer", "subtask": "lock the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7ed9de6021b3542af9905ef5.jpg"}, {"type": "image", "uri": "media/robovqa/fb693b2abf8198c286c89ab9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0230/robovqa_exec_success", "group_id": "robovqa/230", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 230, "taskvar": "230", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "place candy on table", "subtask": "place candy on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f3698f6107b16ed9ea3f6702.jpg"}, {"type": "image", "uri": "media/robovqa/a410efd3ea90b124493508dd.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0231/robovqa_exec_success", "group_id": "robovqa/231", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 231, "taskvar": "231", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Move away from the hanger stand", "subtask": "Move away from the hanger stand", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2886cc559e0712a572a19eac.jpg"}, {"type": "image", "uri": "media/robovqa/a39d495afb567c7e93ad605a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0232/robovqa_exec_success", "group_id": "robovqa/232", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 232, "taskvar": "232", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "move towards the trolley", "subtask": "move towards the trolley", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d7952103ad7f2a3f53cb7179.jpg"}, {"type": "image", "uri": "media/robovqa/50192636f291a48e92802013.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0233/robovqa_exec_success", "group_id": "robovqa/233", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 233, "taskvar": "233", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "hang the cap to the hanger", "subtask": "hang the cap to the hanger", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5b7c9d9a6c05e31530e26162.jpg"}, {"type": "image", "uri": "media/robovqa/62f5184f203c21b2a38a7e41.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0234/robovqa_exec_success", "group_id": "robovqa/234", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 234, "taskvar": "234", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place oat yogurt on table", "subtask": "place oat yogurt on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/538d708332c488f053fdf28a.jpg"}, {"type": "image", "uri": "media/robovqa/7e0829805cad7638acdf4e48.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0235/robovqa_exec_success", "group_id": "robovqa/235", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 235, "taskvar": "235", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "pour cereal into paper cup", "subtask": "pour cereal into paper cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b51dec0b7268ac2761902651.jpg"}, {"type": "image", "uri": "media/robovqa/4c1de174fa80fbff207c492d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0236/robovqa_exec_success", "group_id": "robovqa/236", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 236, "taskvar": "236", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put the oddly cup beside jar", "subtask": "put the oddly cup beside jar", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/1ad7e89eea6c2dda9c3ab0ae.jpg"}, {"type": "image", "uri": "media/robovqa/2ba2d052b151ba6ab072087f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0237/robovqa_exec_success", "group_id": "robovqa/237", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 237, "taskvar": "237", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Move towards the cap", "subtask": "Move towards the cap", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/e976d586cd2fd36e44b1cec9.jpg"}, {"type": "image", "uri": "media/robovqa/d15c5c9f53fae974e2baeb81.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0238/robovqa_exec_success", "group_id": "robovqa/238", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 238, "taskvar": "238", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pick up a green yogurt cup", "subtask": "pick up a green yogurt cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/547417747350939482fbe225.jpg"}, {"type": "image", "uri": "media/robovqa/e0b6f1fd9a1bb59d0b67d02f.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0239/robovqa_exec_success", "group_id": "robovqa/239", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 239, "taskvar": "239", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the pistachios in the glass", "subtask": "put the pistachios in the glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/979526c21aac0c077ea56601.jpg"}, {"type": "image", "uri": "media/robovqa/df2c75067be97681c4117197.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0240/robovqa_exec_success", "group_id": "robovqa/240", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 240, "taskvar": "240", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "unstack the glass", "subtask": "unstack the glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/24dba0ef2c8e079bfb670849.jpg"}, {"type": "image", "uri": "media/robovqa/0d3379de6fe01f1f47203e74.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0241/robovqa_exec_success", "group_id": "robovqa/241", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 241, "taskvar": "241", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pic up the glass", "subtask": "pic up the glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5cf1587cd82931c659aeb8ee.jpg"}, {"type": "image", "uri": "media/robovqa/8c3616c3dafe97380c8c1da6.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0242/robovqa_exec_success", "group_id": "robovqa/242", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 242, "taskvar": "242", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Put the bottle in the dustbin", "subtask": "Put the bottle in the dustbin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c668b4d75f68e157c5c1e7a3.jpg"}, {"type": "image", "uri": "media/robovqa/21938b0264a150c8ac933591.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0243/robovqa_exec_success", "group_id": "robovqa/243", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 243, "taskvar": "243", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the lid in the dustbin", "subtask": "Put the lid in the dustbin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3e8e8bcf78d0b3e6c4243702.jpg"}, {"type": "image", "uri": "media/robovqa/2d0d12033f628fa840510d77.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0244/robovqa_exec_success", "group_id": "robovqa/244", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 244, "taskvar": "244", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the bottle in the dustbin", "subtask": "Put the bottle in the dustbin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/303169758bb5de058f410bfc.jpg"}, {"type": "image", "uri": "media/robovqa/15aac269f52bbc2bebecc3ed.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0245/robovqa_exec_success", "group_id": "robovqa/245", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 245, "taskvar": "245", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the lid in the dustbin", "subtask": "Put the lid in the dustbin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/456a809b507c942451b576ee.jpg"}, {"type": "image", "uri": "media/robovqa/9fbdf8e852b6c06152fbf7e5.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0246/robovqa_exec_success", "group_id": "robovqa/246", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 246, "taskvar": "246", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the lid on the jar", "subtask": "put the lid on the jar", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/98c2da913b132c83048159a6.jpg"}, {"type": "image", "uri": "media/robovqa/9a2a9b72fa168783bda57dcd.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0247/robovqa_exec_success", "group_id": "robovqa/247", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 247, "taskvar": "247", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Go straight", "subtask": "Go straight", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/773dfe8868bbe33515593938.jpg"}, {"type": "image", "uri": "media/robovqa/edc9c726fad31bb745a43711.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0248/robovqa_exec_success", "group_id": "robovqa/248", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 248, "taskvar": "248", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 226, "role": "outcome"}}
{"state": {"task": "Drop the banana in the bowl", "subtask": "Drop the banana in the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/af48127ad119d284efa57a7e.jpg"}, {"type": "image", "uri": "media/robovqa/3267e393f64d7cca4fe1d970.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0249/robovqa_exec_success", "group_id": "robovqa/249", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 249, "taskvar": "249", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Pick up the banana", "subtask": "Pick up the banana", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7371ef9eac6e4b8f41ab969b.jpg"}, {"type": "image", "uri": "media/robovqa/5dcdc2e3ceeb8dac4a9b0196.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0250/robovqa_exec_success", "group_id": "robovqa/250", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 250, "taskvar": "250", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Put the pip corn packet in the tray", "subtask": "Put the pip corn packet in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/ded60853a1c769898e2fb830.jpg"}, {"type": "image", "uri": "media/robovqa/d3b6f418287a094c18c2f38d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0251/robovqa_exec_success", "group_id": "robovqa/251", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 251, "taskvar": "251", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Put the nacho packet in the tray", "subtask": "Put the nacho packet in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/51d50ea14362b5abb7dbfe1c.jpg"}, {"type": "image", "uri": "media/robovqa/71d652435fc74a91f2f9f4be.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0252/robovqa_exec_success", "group_id": "robovqa/252", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 252, "taskvar": "252", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Exit the room", "subtask": "Exit the room", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8a60cb2216efc7017888ba86.jpg"}, {"type": "image", "uri": "media/robovqa/cce45763da66d0686541cf5a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0253/robovqa_exec_success", "group_id": "robovqa/253", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 253, "taskvar": "253", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "Put the nacho packet in the tray", "subtask": "Put the nacho packet in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8dc34f7be57ba92604743d69.jpg"}, {"type": "image", "uri": "media/robovqa/ba2413d82e371325e160d29d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0254/robovqa_exec_success", "group_id": "robovqa/254", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 254, "taskvar": "254", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Put the nacho packet in the tray", "subtask": "Put the nacho packet in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f829520e2d961fb1e6073c22.jpg"}, {"type": "image", "uri": "media/robovqa/34b7e4754bdc87f5d293572c.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0255/robovqa_exec_success", "group_id": "robovqa/255", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 255, "taskvar": "255", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "place brown packet on the counter", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5dc2afe3c9cea9da8f479571.jpg"}, {"type": "image", "uri": "media/robovqa/c6e328d86fc852f3c96cc3b4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0256/robovqa_exec_success", "group_id": "robovqa/256", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 256, "taskvar": "256", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the packet on the table", "subtask": "place the packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/1949940efe58a01877e6e511.jpg"}, {"type": "image", "uri": "media/robovqa/f72872abcb3325afbcbaa4dd.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0257/robovqa_exec_success", "group_id": "robovqa/257", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 257, "taskvar": "257", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick the pen from the table", "subtask": "pick the pen from the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/53752826f5ffc7ee3c9cdf4a.jpg"}, {"type": "image", "uri": "media/robovqa/53752826f5ffc7ee3c9cdf4a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0258/robovqa_exec_success", "group_id": "robovqa/258", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 258, "taskvar": "258", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place brown packet on the counter", "subtask": "place brown packet on the counter", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6360aa25766270688b84d385.jpg"}, {"type": "image", "uri": "media/robovqa/98a7cead45aaaa7c4eefb490.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0259/robovqa_exec_success", "group_id": "robovqa/259", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 259, "taskvar": "259", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the brown packet on the table", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/eb7af660fc87ca532d5ae1be.jpg"}, {"type": "image", "uri": "media/robovqa/d433c19170c91b3cd3b38d02.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0260/robovqa_exec_success", "group_id": "robovqa/260", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 260, "taskvar": "260", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pick the pen from the table", "subtask": "pick the pen from the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fc92e3d6752464e329f3dc51.jpg"}, {"type": "image", "uri": "media/robovqa/31a0ebaff5d6710783bf1373.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0261/robovqa_exec_success", "group_id": "robovqa/261", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 261, "taskvar": "261", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick the pen from the table", "subtask": "pick the pen from the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d60b284dadaaf46aaebee14b.jpg"}, {"type": "image", "uri": "media/robovqa/8b03b6c5716651e40607a3b8.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0262/robovqa_exec_success", "group_id": "robovqa/262", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 262, "taskvar": "262", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick the pen from the table", "subtask": "pick the pen from the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c8fee84dee9888370bdb7594.jpg"}, {"type": "image", "uri": "media/robovqa/2537ad88f0f085600e7abcd6.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0263/robovqa_exec_success", "group_id": "robovqa/263", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 263, "taskvar": "263", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the green chips bag on the top of the drawer", "subtask": "Put the green chips bag on the top of the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c42f91c3d6fbc8f1e9ecb7c8.jpg"}, {"type": "image", "uri": "media/robovqa/297a0395ac09f95ad740a78e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0264/robovqa_exec_success", "group_id": "robovqa/264", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 264, "taskvar": "264", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 244, "role": "outcome"}}
{"state": {"task": "Put brown chips bag on the table", "subtask": "Put brown chips bag on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2452c3ebab101e0fde58e1a0.jpg"}, {"type": "image", "uri": "media/robovqa/e2129f74508c60112959e72b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0265/robovqa_exec_success", "group_id": "robovqa/265", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 265, "taskvar": "265", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the green clour kettie packet on the table", "subtask": "put the green clour kettie packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/741d3e7c8f429590aac4aa31.jpg"}, {"type": "image", "uri": "media/robovqa/13f80e305f22db45097b7d2a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0266/robovqa_exec_success", "group_id": "robovqa/266", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 266, "taskvar": "266", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 246, "role": "outcome"}}
{"state": {"task": "Move the lamp in the left side", "subtask": "Move the lamp in the left side", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4e9b5a93afc24c970c2bed4a.jpg"}, {"type": "image", "uri": "media/robovqa/55c5aad9ee11561f6434fb74.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0267/robovqa_exec_success", "group_id": "robovqa/267", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 267, "taskvar": "267", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the toy on the desk", "subtask": "Put the toy on the desk", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b0ea0cc13ef4e15db657e073.jpg"}, {"type": "image", "uri": "media/robovqa/3137fe46c2f4e317df44bec2.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0268/robovqa_exec_success", "group_id": "robovqa/268", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 268, "taskvar": "268", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Move the lamp forward", "subtask": "Move the lamp forward", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3184766ce26154e95fb3b94a.jpg"}, {"type": "image", "uri": "media/robovqa/73a340f31caea6292afee761.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0269/robovqa_exec_success", "group_id": "robovqa/269", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 269, "taskvar": "269", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "put the blue kettie packet on the table", "subtask": "put the blue kettie packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/61543d9cb164d1fe8124eeed.jpg"}, {"type": "image", "uri": "media/robovqa/f31e6a61148b3d9872a5783a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0270/robovqa_exec_success", "group_id": "robovqa/270", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 270, "taskvar": "270", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 242, "role": "outcome"}}
{"state": {"task": "keep pen on trolley", "subtask": "keep pen on trolley", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/51fcf6cab2029e2d4e6a0a4b.jpg"}, {"type": "image", "uri": "media/robovqa/51978920ede2592ad7131163.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0271/robovqa_exec_success", "group_id": "robovqa/271", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 271, "taskvar": "271", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "Place the pen in the cart", "subtask": "Place the pen in the cart", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/041fb838fe36a8ac2bc373c1.jpg"}, {"type": "image", "uri": "media/robovqa/d4525e0335941370bb56ac0b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0272/robovqa_exec_success", "group_id": "robovqa/272", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 272, "taskvar": "272", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Place the stickey note book in the cart", "subtask": "Place the stickey note book in the cart", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fab23a52a2f83f6287ebe9b0.jpg"}, {"type": "image", "uri": "media/robovqa/f03a4bbd3ab3eb9d726a6d85.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0273/robovqa_exec_success", "group_id": "robovqa/273", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 273, "taskvar": "273", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "place the yellow packet on the black table", "subtask": "place the yellow packet on the black table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/481849e793b89602e9b8fd86.jpg"}, {"type": "image", "uri": "media/robovqa/1b48f929804d2d8f649d2417.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0274/robovqa_exec_success", "group_id": "robovqa/274", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 274, "taskvar": "274", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "move pen to pen holder", "subtask": "move grasped object near the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/78cb89b93550277881c41f10.jpg"}, {"type": "image", "uri": "media/robovqa/ff8ac4a6b4395b4e7631f2fe.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0275/robovqa_exec_success", "group_id": "robovqa/275", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 275, "taskvar": "275", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 235, "role": "outcome"}}
{"state": {"task": "move the cart to the table", "subtask": "move the cart to the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c57ec88b750969a6b8f73726.jpg"}, {"type": "image", "uri": "media/robovqa/9715f6af4e7dc33a5b8dcd25.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0276/robovqa_exec_success", "group_id": "robovqa/276", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 276, "taskvar": "276", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Pick the Pistachio cup from the table", "subtask": "Pick the Pistachio cup from the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7ad84eaa34302b0e6178ec28.jpg"}, {"type": "image", "uri": "media/robovqa/a940ba99e77b012c73daa6b9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0277/robovqa_exec_success", "group_id": "robovqa/277", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 277, "taskvar": "277", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "Adjust the pen holder", "subtask": "Adjust the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8e42d976bd5396d5259d6528.jpg"}, {"type": "image", "uri": "media/robovqa/915846a71de809105c401a0c.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0278/robovqa_exec_success", "group_id": "robovqa/278", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 278, "taskvar": "278", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "Put the Pistachio inside the cup", "subtask": "move the Pistachio from jar to the cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2d9c455f3bdd652993c22d75.jpg"}, {"type": "image", "uri": "media/robovqa/1c4cf1dd7f6c0b786c54e020.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0279/robovqa_exec_success", "group_id": "robovqa/279", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 279, "taskvar": "279", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "Leave the room with the cup", "subtask": "Leave the room with the cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/008c297313efb19bbea4f4bc.jpg"}, {"type": "image", "uri": "media/robovqa/42e2a25070323ec3846d3fab.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0280/robovqa_exec_success", "group_id": "robovqa/280", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 280, "taskvar": "280", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the blue pens in the pen holder", "subtask": "Put the blue pens in the pen holder", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/1b197f75a434f9b8096aeab0.jpg"}, {"type": "image", "uri": "media/robovqa/c5547326ab2358e5796b9e65.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0281/robovqa_exec_success", "group_id": "robovqa/281", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 281, "taskvar": "281", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Upright a cup from the stack", "subtask": "Upright a cup from the stack", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d6fed65e81dcc4171712f8b3.jpg"}, {"type": "image", "uri": "media/robovqa/d6fed65e81dcc4171712f8b3.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0282/robovqa_exec_success", "group_id": "robovqa/282", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 282, "taskvar": "282", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Put the pear into the bowl", "subtask": "Put the pear into the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/44367c712ba7b89ce38880db.jpg"}, {"type": "image", "uri": "media/robovqa/8329a6e56d9268d305adc234.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0283/robovqa_exec_success", "group_id": "robovqa/283", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 283, "taskvar": "283", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the pear into the bowl", "subtask": "Put the pear into the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/389dcbb0240f22906ec3a45a.jpg"}, {"type": "image", "uri": "media/robovqa/9ba13a211a9fc9369d72d39a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0284/robovqa_exec_success", "group_id": "robovqa/284", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 284, "taskvar": "284", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "move second glass towards right", "subtask": "grasp glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b7e2134dcdad9544bf3db7e9.jpg"}, {"type": "image", "uri": "media/robovqa/e42b4868350eacd956c48fe9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0285/robovqa_exec_success", "group_id": "robovqa/285", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 285, "taskvar": "285", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "thow the green glass between the four glasses", "subtask": "thow the green glass between the four glasses", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8b064f2690ae08ebcc2904ad.jpg"}, {"type": "image", "uri": "media/robovqa/76f8e431c095b2e1f6674051.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0286/robovqa_exec_success", "group_id": "robovqa/286", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 286, "taskvar": "286", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "drop the pear into the bowl", "subtask": "drop the pear into the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/389dcbb0240f22906ec3a45a.jpg"}, {"type": "image", "uri": "media/robovqa/3c598ee4b1a3de41b1ac54ad.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0287/robovqa_exec_success", "group_id": "robovqa/287", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 287, "taskvar": "287", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the pear into the bowl", "subtask": "Put the pear into the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/911e6daca56775035fe04a39.jpg"}, {"type": "image", "uri": "media/robovqa/80ce087abf1f9f5639eb9c7b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0288/robovqa_exec_success", "group_id": "robovqa/288", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 288, "taskvar": "288", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick up the green glass", "subtask": "pick up the green glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/97c30ab9236a590fcf026292.jpg"}, {"type": "image", "uri": "media/robovqa/ef8d98788e53231a820fd012.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0289/robovqa_exec_success", "group_id": "robovqa/289", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 289, "taskvar": "289", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "thow the green glass between the four glasses", "subtask": "thow the green glass between the four glasses", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/a3ece2cd5fc51b1c496dff49.jpg"}, {"type": "image", "uri": "media/robovqa/c0fbc927f0914fd6b7567eff.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0290/robovqa_exec_success", "group_id": "robovqa/290", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 290, "taskvar": "290", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "put the packets that are in the drawer into the bowl", "subtask": "put the packets that are in the drawer into the bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/109403092108b73823cab464.jpg"}, {"type": "image", "uri": "media/robovqa/38f4a84e4fb7c7e4946b01fe.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0291/robovqa_exec_success", "group_id": "robovqa/291", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 291, "taskvar": "291", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 244, "role": "outcome"}}
{"state": {"task": "press the second button", "subtask": "press the second button", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/57b8190af8fe1852e56c9ab9.jpg"}, {"type": "image", "uri": "media/robovqa/91187ba1bc72e7c54349db93.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0292/robovqa_exec_success", "group_id": "robovqa/292", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 292, "taskvar": "292", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pull the top drawer", "subtask": "pull the top drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/09211209768c9d4eca379c30.jpg"}, {"type": "image", "uri": "media/robovqa/09211209768c9d4eca379c30.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0293/robovqa_exec_success", "group_id": "robovqa/293", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 293, "taskvar": "293", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "open the drawer", "subtask": "open the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/9e2a59ca599b7e2f115e14f8.jpg"}, {"type": "image", "uri": "media/robovqa/af338c54ff178fcbec37cf5b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0294/robovqa_exec_success", "group_id": "robovqa/294", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 294, "taskvar": "294", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/5a8c61f5dbad63efa6028811.jpg"}, {"type": "image", "uri": "media/robovqa/a34293cb1711d5a1b6928505.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0295/robovqa_exec_success", "group_id": "robovqa/295", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 295, "taskvar": "295", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "adjust the glass", "subtask": "grasp the glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2db2b1804d85f11ad30fe7d2.jpg"}, {"type": "image", "uri": "media/robovqa/0af100ea79a7c42b2a3b3306.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0296/robovqa_exec_success", "group_id": "robovqa/296", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 296, "taskvar": "296", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 229, "role": "outcome"}}
{"state": {"task": "place the glass on table", "subtask": "grasp the glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4c20fe3b61e54daa07a3c2d8.jpg"}, {"type": "image", "uri": "media/robovqa/5e8939e33c756e6ea1ae6561.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0297/robovqa_exec_success", "group_id": "robovqa/297", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 297, "taskvar": "297", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "place the glass on table", "subtask": "place the glass on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4cff1c364254093e5459a220.jpg"}, {"type": "image", "uri": "media/robovqa/4da8e171d1a51c7731e9940b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0298/robovqa_exec_success", "group_id": "robovqa/298", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 298, "taskvar": "298", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/399c0e4bfc4f0953ee81dbe5.jpg"}, {"type": "image", "uri": "media/robovqa/fff4bb909a85705f6d943523.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0299/robovqa_exec_success", "group_id": "robovqa/299", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 299, "taskvar": "299", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "drop brown box on the left table", "subtask": "drop brown box on the left table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/67afec5c87145cd2eed07935.jpg"}, {"type": "image", "uri": "media/robovqa/eaffa019f2c19e692bff2728.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0300/robovqa_exec_success", "group_id": "robovqa/300", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 300, "taskvar": "300", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put toy on the table", "subtask": "put toy on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/02f73e0cf1a7e7f7b8fe0ccf.jpg"}, {"type": "image", "uri": "media/robovqa/4ee59847a35b581f7e957427.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0301/robovqa_exec_success", "group_id": "robovqa/301", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 301, "taskvar": "301", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "drop brown box on the left table", "subtask": "drop brown box on the left table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3517498467bc424b64465b7d.jpg"}, {"type": "image", "uri": "media/robovqa/fef44e8dc41dcc918f744089.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0302/robovqa_exec_success", "group_id": "robovqa/302", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 302, "taskvar": "302", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put toy on the table", "subtask": "put toy on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f12d8fab912151a255f4eef9.jpg"}, {"type": "image", "uri": "media/robovqa/4598f2cd02ed23bd88b2bc50.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0303/robovqa_exec_success", "group_id": "robovqa/303", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 303, "taskvar": "303", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put toy on the table", "subtask": "put toy on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/0330ef23e2659098c1000a80.jpg"}, {"type": "image", "uri": "media/robovqa/ca43ac08fc3829c3ff014e51.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0304/robovqa_exec_success", "group_id": "robovqa/304", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 304, "taskvar": "304", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put toy on the table", "subtask": "put toy on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/aea2590f0378766986029f30.jpg"}, {"type": "image", "uri": "media/robovqa/2e7938df6b140b1e2ae5930b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0305/robovqa_exec_success", "group_id": "robovqa/305", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 305, "taskvar": "305", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "close the first compartment of the drawer", "subtask": "close the first compartment of the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/aebd02286774216b0ddd0520.jpg"}, {"type": "image", "uri": "media/robovqa/b481ddd96a75d9244ea07871.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0306/robovqa_exec_success", "group_id": "robovqa/306", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 306, "taskvar": "306", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Cross the threshold of the cabin, and enter", "subtask": "Cross the threshold of the cabin, and enter", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/7dc2b473067f2ecf021d6f16.jpg"}, {"type": "image", "uri": "media/robovqa/981c47dcb3c056293fe8fbb4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0307/robovqa_exec_success", "group_id": "robovqa/307", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 307, "taskvar": "307", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "place the packet in the drawer", "subtask": "place the packet in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6b16ba4917b97417b3a19d71.jpg"}, {"type": "image", "uri": "media/robovqa/8c06fb3cc4a8f8f01127b9aa.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0308/robovqa_exec_success", "group_id": "robovqa/308", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 308, "taskvar": "308", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Exit the cabin", "subtask": "Exit the cabin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6d47d3417e0679e4f9645f1a.jpg"}, {"type": "image", "uri": "media/robovqa/dc85f899e80a9138cc11d1c8.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0309/robovqa_exec_success", "group_id": "robovqa/309", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 309, "taskvar": "309", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "place the packet in the drawer", "subtask": "place the packet in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/2ac4e50db4f7260b7e215960.jpg"}, {"type": "image", "uri": "media/robovqa/a094ef2e19f44b369ebc8ded.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0310/robovqa_exec_success", "group_id": "robovqa/310", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 310, "taskvar": "310", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Place the cup inside the green dustbin", "subtask": "release", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fc01216c9483cd9448efc5a6.jpg"}, {"type": "image", "uri": "media/robovqa/ce197f36166a305f23d66cdc.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0311/robovqa_exec_success", "group_id": "robovqa/311", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 311, "taskvar": "311", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 231, "role": "outcome"}}
{"state": {"task": "place all the packets in the drawer", "subtask": "place all the packets in the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/bb61298fcc92f259687d29ae.jpg"}, {"type": "image", "uri": "media/robovqa/8b27b0c9408ce263ee513f0e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0312/robovqa_exec_success", "group_id": "robovqa/312", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 312, "taskvar": "312", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Place the cup inside the green dustbin", "subtask": "Place the cup inside the green dustbin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/731abebbac9985776c781bb0.jpg"}, {"type": "image", "uri": "media/robovqa/938ddb1b73917cd2aa5e30c8.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0313/robovqa_exec_success", "group_id": "robovqa/313", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 313, "taskvar": "313", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "Pull the Box out of the shelves", "subtask": "grasp box", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d74645e924d4bce7a654faf5.jpg"}, {"type": "image", "uri": "media/robovqa/00ef4da0c1f567b61bfc9482.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0314/robovqa_exec_success", "group_id": "robovqa/314", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 314, "taskvar": "314", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "take the billiards ball towards the billiards table", "subtask": "take the billiards ball towards the billiards table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8ca886ebcad7c7ed078ca57b.jpg"}, {"type": "image", "uri": "media/robovqa/2088861bbff220adae5ed0ba.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0315/robovqa_exec_success", "group_id": "robovqa/315", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 315, "taskvar": "315", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 242, "role": "outcome"}}
{"state": {"task": "place the bottle in the recycle bin", "subtask": "place the bottle in the recycle bin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/97f2b910acc384cc6661bdf8.jpg"}, {"type": "image", "uri": "media/robovqa/eb4fab68f1fc847e154933d9.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0316/robovqa_exec_success", "group_id": "robovqa/316", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 316, "taskvar": "316", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the bottle in the recycle bin", "subtask": "place the bottle in the recycle bin", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/29911337b0a70b707fb1ae7a.jpg"}, {"type": "image", "uri": "media/robovqa/b42b78e7f6a8961098fcef73.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0317/robovqa_exec_success", "group_id": "robovqa/317", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 317, "taskvar": "317", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the box out from drawer", "subtask": "put the box out from drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/26035d9f8f0c9fe6a1925290.jpg"}, {"type": "image", "uri": "media/robovqa/26035d9f8f0c9fe6a1925290.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0318/robovqa_exec_success", "group_id": "robovqa/318", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 318, "taskvar": "318", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Pull the Box out of the shelves", "subtask": "Pull the Box out of the shelves", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6fdd2907b2ebbb6a40889d9c.jpg"}, {"type": "image", "uri": "media/robovqa/eeb6128e53533ab03dcaa80e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0319/robovqa_exec_success", "group_id": "robovqa/319", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 319, "taskvar": "319", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the orange that is on the table into the white bowl with the banana and the other orange", "subtask": "move orange into the white bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/fbf44e6b563c7295ddcab0fd.jpg"}, {"type": "image", "uri": "media/robovqa/414395dc5301335cee6e4621.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0320/robovqa_exec_success", "group_id": "robovqa/320", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 320, "taskvar": "320", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 247, "role": "outcome"}}
{"state": {"task": "remove wooden piece from the left drawer and put it on the table", "subtask": "move grasped wooden object out of the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/164046563c2e8b698158efdb.jpg"}, {"type": "image", "uri": "media/robovqa/f72b1c60b034cb79179034d3.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0321/robovqa_exec_success", "group_id": "robovqa/321", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 321, "taskvar": "321", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 244, "role": "outcome"}}
{"state": {"task": "put circle in the drawer set", "subtask": "put circle in the drawer set", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/255a998732de9c5490f2d855.jpg"}, {"type": "image", "uri": "media/robovqa/c3cf0d1a6cdca02a1e37749e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0322/robovqa_exec_success", "group_id": "robovqa/322", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 322, "taskvar": "322", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Put the candy in the brown bowl", "subtask": "Put the candy in the brown bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/4bbacb04734a5f04c1b8154a.jpg"}, {"type": "image", "uri": "media/robovqa/198623e9f4b39b3b57c74c55.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0323/robovqa_exec_success", "group_id": "robovqa/323", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 323, "taskvar": "323", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "label_conflict": "reward 1 but Guardian category 'no_progress'; label follows the reward (= Guardian's gold True/False answer)", "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/be5e43b67741688324234df7.jpg"}, {"type": "image", "uri": "media/robovqa/62d44264455b5b2d93de5d27.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0324/robovqa_exec_success", "group_id": "robovqa/324", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 324, "taskvar": "324", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "close the drawer", "subtask": "close the drawer", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/12b5499d98f20f56823a5f59.jpg"}, {"type": "image", "uri": "media/robovqa/6bff6e857e7f3dc646f4376a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0325/robovqa_exec_success", "group_id": "robovqa/325", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 325, "taskvar": "325", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 228, "role": "outcome"}}
{"state": {"task": "put orange in the white bowl", "subtask": "put orange in the white bowl", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/68dcc4a3a234bde69e7ef15e.jpg"}, {"type": "image", "uri": "media/robovqa/68dcc4a3a234bde69e7ef15e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0326/robovqa_exec_success", "group_id": "robovqa/326", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 326, "taskvar": "326", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the brown packet in the basket", "subtask": "put the brown packet in the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/a442a6c59374eb74dbe78a7c.jpg"}, {"type": "image", "uri": "media/robovqa/5cf5783f5828765c47e04d08.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0327/robovqa_exec_success", "group_id": "robovqa/327", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 327, "taskvar": "327", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the scoop in the stand", "subtask": "place the scoop in the stand", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d1094335bdab62cf77f7badc.jpg"}, {"type": "image", "uri": "media/robovqa/c79206d0f6ca9a203ad08f4e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0328/robovqa_exec_success", "group_id": "robovqa/328", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 328, "taskvar": "328", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place pistachios in the cup", "subtask": "place pistachios in the cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6bba9aedb8f81b4354ad1cea.jpg"}, {"type": "image", "uri": "media/robovqa/eca24e6a3c09f20357fc23d4.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0329/robovqa_exec_success", "group_id": "robovqa/329", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 329, "taskvar": "329", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the lid of jar on table", "subtask": "place the lid of jar on table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3b669c1e8cc6aae6c0198210.jpg"}, {"type": "image", "uri": "media/robovqa/8f26b8c4b11366f0d9ff9305.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0330/robovqa_exec_success", "group_id": "robovqa/330", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 330, "taskvar": "330", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the brown packet in the basket", "subtask": "put the brown packet in the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/aeba8605ed86c1c960700032.jpg"}, {"type": "image", "uri": "media/robovqa/87e622f6b2a251dea5214178.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0331/robovqa_exec_success", "group_id": "robovqa/331", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 331, "taskvar": "331", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the blue packet in the basket", "subtask": "put the blue packet in the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/3a7c56656f65692764f29288.jpg"}, {"type": "image", "uri": "media/robovqa/d4e884910cc544ef1241d59e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0332/robovqa_exec_success", "group_id": "robovqa/332", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 332, "taskvar": "332", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put the blue packet in the tray", "subtask": "put the blue packet in the tray", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6f9f5448b601d61cc0b9aad8.jpg"}, {"type": "image", "uri": "media/robovqa/0800d5f5e5297873dd876bff.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0333/robovqa_exec_success", "group_id": "robovqa/333", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 333, "taskvar": "333", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place pistachios in the cup", "subtask": "place pistachios in the cup", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/df73d5186b4993d934074bda.jpg"}, {"type": "image", "uri": "media/robovqa/87a399480c393b2016015861.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0334/robovqa_exec_success", "group_id": "robovqa/334", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 334, "taskvar": "334", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "Put the marker in the basket", "subtask": "Put the marker in the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/1ea9a2313a6f1c7cf8ddeb10.jpg"}, {"type": "image", "uri": "media/robovqa/a892f56ffb84d6f295edbe79.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0335/robovqa_exec_success", "group_id": "robovqa/335", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 335, "taskvar": "335", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put the pens into the trolley", "subtask": "move grasped pen to black trolley", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/790ec2aa8c7051dd65a146ce.jpg"}, {"type": "image", "uri": "media/robovqa/ec6609a46fd90718b5916780.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0336/robovqa_exec_success", "group_id": "robovqa/336", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 336, "taskvar": "336", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 237, "role": "outcome"}}
{"state": {"task": "put two pears on the table", "subtask": "put two pears on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/318b5de7077ee342df1b4a9d.jpg"}, {"type": "image", "uri": "media/robovqa/4e1be65955954ce642d4baad.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0337/robovqa_exec_success", "group_id": "robovqa/337", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 337, "taskvar": "337", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "put apple on the table", "subtask": "put apple on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/58cd5cae57c83dcb8a3d306e.jpg"}, {"type": "image", "uri": "media/robovqa/281b863ec99685b7a244d384.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0338/robovqa_exec_success", "group_id": "robovqa/338", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 338, "taskvar": "338", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "put pear on the table", "subtask": "put pear on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/844bb7d8c96c412365cdcf9f.jpg"}, {"type": "image", "uri": "media/robovqa/4900225a8302fd42edff652b.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0339/robovqa_exec_success", "group_id": "robovqa/339", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 339, "taskvar": "339", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 232, "role": "outcome"}}
{"state": {"task": "pickup the black object", "subtask": "pickup the black object", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/a40694ff39eab1fa1c837e15.jpg"}, {"type": "image", "uri": "media/robovqa/27b112b232e9d94c4c03b409.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0340/robovqa_exec_success", "group_id": "robovqa/340", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 340, "taskvar": "340", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "put the pen in the basket", "subtask": "put the pen in the basket", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/d7eee028800495d364d61952.jpg"}, {"type": "image", "uri": "media/robovqa/d7eee028800495d364d61952.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0341/robovqa_exec_success", "group_id": "robovqa/341", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 341, "taskvar": "341", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place the two glasses on the table", "subtask": "place the two glasses on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f820c22b775b64137eb4e602.jpg"}, {"type": "image", "uri": "media/robovqa/9d1d772b4a691fd9578d7e6e.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0342/robovqa_exec_success", "group_id": "robovqa/342", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 342, "taskvar": "342", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "erase the white board", "subtask": "erase the white board", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/371fac3a3c273bad9fa8d2e3.jpg"}, {"type": "image", "uri": "media/robovqa/1a6aa3f64605e90b10337904.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0343/robovqa_exec_success", "group_id": "robovqa/343", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 343, "taskvar": "343", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "pick up the eraser sponge", "subtask": "pick up the eraser sponge", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/f5dc0383c63b9a0dccdaef01.jpg"}, {"type": "image", "uri": "media/robovqa/8572f19b5bda8d5255572158.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0344/robovqa_exec_success", "group_id": "robovqa/344", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 344, "taskvar": "344", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "place an orange glass on the table", "subtask": "place an orange glass on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/0e1255d3bd94baee935dfa42.jpg"}, {"type": "image", "uri": "media/robovqa/31daeb9760a3fdb1cab9e1c8.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0345/robovqa_exec_success", "group_id": "robovqa/345", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 345, "taskvar": "345", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the eraser sponge on the table", "subtask": "place the eraser sponge on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8ce4d9603bcc5ee1f7083d86.jpg"}, {"type": "image", "uri": "media/robovqa/5ee1100bc202d0bb67915324.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0346/robovqa_exec_success", "group_id": "robovqa/346", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 346, "taskvar": "346", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 238, "role": "outcome"}}
{"state": {"task": "unstack two glasses", "subtask": "unstack two glasses", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/6e05b222c130187fad5e37e4.jpg"}, {"type": "image", "uri": "media/robovqa/0b5ad42a1f3109ae516aa288.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0347/robovqa_exec_success", "group_id": "robovqa/347", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 347, "taskvar": "347", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "unstack the glass", "subtask": "unstack the glass", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/c02728303a2c25d7918cd7d9.jpg"}, {"type": "image", "uri": "media/robovqa/0da81a66a52d2feda65daf04.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0348/robovqa_exec_success", "group_id": "robovqa/348", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 348, "taskvar": "348", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 230, "role": "outcome"}}
{"state": {"task": "put black marker on the table", "subtask": "put black marker on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/08b8387f3c1bc8bae0eac9a4.jpg"}, {"type": "image", "uri": "media/robovqa/a971a37a9d58222774f4b79d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0349/robovqa_exec_success", "group_id": "robovqa/349", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 349, "taskvar": "349", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "put blue pen on the table", "subtask": "put blue pen on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/dfb2881a83955d3c47974c41.jpg"}, {"type": "image", "uri": "media/robovqa/0aea13487ad16e40a07fbf0d.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0350/robovqa_exec_success", "group_id": "robovqa/350", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 350, "taskvar": "350", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "pick up the watermelon hint bottle", "subtask": "pick up the watermelon hint bottle", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b7d6b3324ce5e78c40865ea5.jpg"}, {"type": "image", "uri": "media/robovqa/d6ebbc50835a79d4f60d5a01.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0351/robovqa_exec_success", "group_id": "robovqa/351", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 351, "taskvar": "351", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "place the watermelon hint bottle onto the table top", "subtask": "place the watermelon hint bottle onto the table top", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/a58ea59955e6322965072a2d.jpg"}, {"type": "image", "uri": "media/robovqa/5eba9ac6f6c309d2b5d9cf12.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0352/robovqa_exec_success", "group_id": "robovqa/352", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 352, "taskvar": "352", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 242, "role": "outcome"}}
{"state": {"task": "put the watermelon hint bottle on the table", "subtask": "put the watermelon hint bottle on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b578ceb485769a313db5e81b.jpg"}, {"type": "image", "uri": "media/robovqa/356c29924a9a1dc428b4adb6.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0353/robovqa_exec_success", "group_id": "robovqa/353", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 353, "taskvar": "353", "episode_id": null, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 240, "role": "outcome"}}
{"state": {"task": "put the glass into the trash", "subtask": "put the glass into the trash", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/688ac759f379c7681c989657.jpg"}, {"type": "image", "uri": "media/robovqa/5e4fe5b6137fc86ff65ffa0a.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0354/robovqa_exec_success", "group_id": "robovqa/354", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 354, "taskvar": "354", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[960, 540], [960, 540]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 234, "role": "outcome"}}
{"state": {"task": "Place the nacho packet from the tray onto the table", "subtask": "Place the nacho packet from the tray onto the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/b12181ae6ce646d7f5699cc9.jpg"}, {"type": "image", "uri": "media/robovqa/79a11707b1dd238d04cf3dd6.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0355/robovqa_exec_success", "group_id": "robovqa/355", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 355, "taskvar": "355", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 244, "role": "outcome"}}
{"state": {"task": "Place the popcorn packet on the table", "subtask": "Place the popcorn packet on the table", "images": ["start of the subtask", "end of the subtask"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a robot or a human (RoboVQA mixes an Everyday Robots mobile manipulator, a human arm and a human using a grasping tool). The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from one camera at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "robovqa_exec_success"}}, "media": [{"type": "image", "uri": "media/robovqa/8f66b8d2ea920b63e54310d4.jpg"}, {"type": "image", "uri": "media/robovqa/f4cd207dc777c4bae1cd02bf.jpg"}], "_meta": {"source": "robovqa", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "robovqa/exec0356/robovqa_exec_success", "group_id": "robovqa/356", "family": "robovqa_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "RoboVQA (execution-success subset)", "native_line": 356, "taskvar": "356", "episode_id": null, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[640, 512], [640, 512]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 2, "approx_tokens": 236, "role": "outcome"}}
{"state": {"task": "pick up the fruit, and lift it up, then put it back on the table, then put it in the on the floor", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ad4501b9673211a57736fdcf.png"}, {"type": "image", "uri": "media/ur5fail/5ac9c43e9ffe6922926f8495.png"}, {"type": "image", "uri": "media/ur5fail/41a6fd56ee1c087689219044.png"}, {"type": "image", "uri": "media/ur5fail/d10db5453792507257c1e727.png"}, {"type": "image", "uri": "media/ur5fail/7c704973d6594c0403eac985.png"}, {"type": "image", "uri": "media/ur5fail/b5f4b7773eb634d1928b22a5.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0000/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_pick_tool+0/173817592039", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 0, "taskvar": "real_pick_tool+0", "episode_id": 173817592039, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 359, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box, then move out the tuna can and put it back on table, and put the tuna can on the floor", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/2b313fa95f6d493bc05b8185.png"}, {"type": "image", "uri": "media/ur5fail/d3b0be87528ac433ae5c31e2.png"}, {"type": "image", "uri": "media/ur5fail/e6f5e3aed9ca9587f282929f.png"}, {"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0001/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 1, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "pick up the ball", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/f349279bbeb2818226a25b3f.png"}, {"type": "image", "uri": "media/ur5fail/542104e9c1edbe8526c95876.png"}, {"type": "image", "uri": "media/ur5fail/d20ad84fb548022298500623.png"}, {"type": "image", "uri": "media/ur5fail/59762efe252de42dc815e8f6.png"}, {"type": "image", "uri": "media/ur5fail/61e702940222756c349615a6.png"}, {"type": "image", "uri": "media/ur5fail/6ae940908fac4de2519127fa.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0002/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_pick_tool+0/173817611377", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 2, "taskvar": "real_pick_tool+0", "episode_id": 173817611377, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 337, "role": "error_localization"}}
{"state": {"task": "grasp the lightest object", "subtask": "grasp the lightest object", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/f349279bbeb2818226a25b3f.png"}, {"type": "image", "uri": "media/ur5fail/542104e9c1edbe8526c95876.png"}, {"type": "image", "uri": "media/ur5fail/d20ad84fb548022298500623.png"}, {"type": "image", "uri": "media/ur5fail/59762efe252de42dc815e8f6.png"}, {"type": "image", "uri": "media/ur5fail/61e702940222756c349615a6.png"}, {"type": "image", "uri": "media/ur5fail/6ae940908fac4de2519127fa.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0003/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_pick_tool+0/173817611377", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 3, "taskvar": "real_pick_tool+0", "episode_id": 173817611377, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "pick up the fruit, and lift it up, then put it back on the table", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ad4501b9673211a57736fdcf.png"}, {"type": "image", "uri": "media/ur5fail/5ac9c43e9ffe6922926f8495.png"}, {"type": "image", "uri": "media/ur5fail/41a6fd56ee1c087689219044.png"}, {"type": "image", "uri": "media/ur5fail/d10db5453792507257c1e727.png"}, {"type": "image", "uri": "media/ur5fail/7c704973d6594c0403eac985.png"}, {"type": "image", "uri": "media/ur5fail/b5f4b7773eb634d1928b22a5.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0006/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_pick_tool+0/173817592039", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 6, "taskvar": "real_pick_tool+0", "episode_id": 173817592039, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 350, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "subtask": "grasp tuna can", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}, {"type": "image", "uri": "media/ur5fail/b14edf6d62c5cf2ed5488855.png"}, {"type": "image", "uri": "media/ur5fail/aad32040564cbe94530b39b8.png"}, {"type": "image", "uri": "media/ur5fail/3bc30872fe8dcd494c0c4859.png"}, {"type": "image", "uri": "media/ur5fail/3258013a066c2f954b29e48d.png"}, {"type": "image", "uri": "media/ur5fail/acea5dc1d3dc6bcac0e38461.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0007/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 7, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 348, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/2b313fa95f6d493bc05b8185.png"}, {"type": "image", "uri": "media/ur5fail/d3b0be87528ac433ae5c31e2.png"}, {"type": "image", "uri": "media/ur5fail/e6f5e3aed9ca9587f282929f.png"}, {"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0009/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 9, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 350, "role": "error_localization"}}
{"state": {"task": "put the tuna can to the pizza slice then put the corn in the box", "subtask": "move grasped object to the pizza slice", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}, {"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}, {"type": "image", "uri": "media/ur5fail/b14edf6d62c5cf2ed5488855.png"}, {"type": "image", "uri": "media/ur5fail/aad32040564cbe94530b39b8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0011/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 11, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "put the tuna can next to the strawberry then the strawberry in the box, then the donut in the box", "subtask": "move grasped object next to strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}, {"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0012/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 12, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 359, "role": "error_localization"}}
{"state": {"task": "put all the fruits into the box", "subtask": "put strawberry into the box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/3bc30872fe8dcd494c0c4859.png"}, {"type": "image", "uri": "media/ur5fail/3258013a066c2f954b29e48d.png"}, {"type": "image", "uri": "media/ur5fail/acea5dc1d3dc6bcac0e38461.png"}, {"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0014/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 14, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the pizza slice in the box", "subtask": "grasp pizza slice", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}, {"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0015/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 15, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 349, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the pizza slice in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}, {"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0016/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 16, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn cob between the strawberry and the donut", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}, {"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}, {"type": "image", "uri": "media/ur5fail/b14edf6d62c5cf2ed5488855.png"}, {"type": "image", "uri": "media/ur5fail/aad32040564cbe94530b39b8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0017/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 17, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 355, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "subtask": "grasp corn", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}, {"type": "image", "uri": "media/ur5fail/70c8c2e550dbfcba2a12a170.png"}, {"type": "image", "uri": "media/ur5fail/4bef3431062e072baa30f95c.png"}, {"type": "image", "uri": "media/ur5fail/f5c0212f6f9862ecf7f6860a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0018/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 18, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 347, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the black grapes in the box", "subtask": "grasp black grapes", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}, {"type": "image", "uri": "media/ur5fail/70c8c2e550dbfcba2a12a170.png"}, {"type": "image", "uri": "media/ur5fail/4bef3431062e072baa30f95c.png"}, {"type": "image", "uri": "media/ur5fail/f5c0212f6f9862ecf7f6860a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0019/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 19, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 349, "role": "error_localization"}}
{"state": {"task": "grasp the corn cob, lift it up, and place it in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}, {"type": "image", "uri": "media/ur5fail/70c8c2e550dbfcba2a12a170.png"}, {"type": "image", "uri": "media/ur5fail/4bef3431062e072baa30f95c.png"}, {"type": "image", "uri": "media/ur5fail/f5c0212f6f9862ecf7f6860a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0021/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 21, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "grasp the tuna can, lift it up, and place it in the box, grasp the corn cob, lift it up, and place it in the box, then put the corn cob back to the table", "subtask": "move grasped object down on the table", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}, {"type": "image", "uri": "media/ur5fail/60d16780c9068bf3e9292bc5.png"}, {"type": "image", "uri": "media/ur5fail/22dce9957f0f5bfc42486227.png"}, {"type": "image", "uri": "media/ur5fail/177a01339f884083162e5454.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0024/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 24, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 380, "role": "error_localization"}}
{"state": {"task": "grasp the tuna can, lift it up, and place it in the box, grasp the corn cob, lift it up, and place it in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_progress", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}, {"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0026/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 26, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 368, "role": "error_localization"}}
{"state": {"task": "grasp the donut, and place it in the box, grasp the corn cob, lift it up, and place it in the box", "subtask": "put the donut in the box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}, {"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0027/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 27, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 365, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the donut in the box", "subtask": "grasp donut", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/56206ecfe1657377c4fe979a.png"}, {"type": "image", "uri": "media/ur5fail/0caec930fea30714f61b6135.png"}, {"type": "image", "uri": "media/ur5fail/d56efaec3b2fdae965bea502.png"}, {"type": "image", "uri": "media/ur5fail/c618787c6df83bd140f0b783.png"}, {"type": "image", "uri": "media/ur5fail/592dbd7a46cc410b1ae6119a.png"}, {"type": "image", "uri": "media/ur5fail/4c391536923d24ac64641d8b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0028/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/1", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 28, "taskvar": "real_put_food_in_box+0", "episode_id": 1, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 349, "role": "error_localization"}}
{"state": {"task": "put the grapes in the box then the banana", "subtask": "grasp grapes", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}, {"type": "image", "uri": "media/ur5fail/9641470ccc34dc3a51bebcd2.png"}, {"type": "image", "uri": "media/ur5fail/5019194395c260a6640fa98a.png"}, {"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0033/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 33, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box then the pear", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}, {"type": "image", "uri": "media/ur5fail/9641470ccc34dc3a51bebcd2.png"}, {"type": "image", "uri": "media/ur5fail/5019194395c260a6640fa98a.png"}, {"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0034/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 34, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "put the banana between yellow peach and strawberry, then put the black grapes in the box", "subtask": "move grasped object to between yellow peach and strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/faa95603cf9dde8e979a631f.png"}, {"type": "image", "uri": "media/ur5fail/128b0efb96293dfa79b0297f.png"}, {"type": "image", "uri": "media/ur5fail/9c91b58f97136cf6a3de32c6.png"}, {"type": "image", "uri": "media/ur5fail/9d1ecd236442311b1e345803.png"}, {"type": "image", "uri": "media/ur5fail/9bc0041695b48ac00f3d1ad4.png"}, {"type": "image", "uri": "media/ur5fail/f3e49635bee479b8c920552f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0038/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 38, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 357, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "grasp lemon", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}, {"type": "image", "uri": "media/ur5fail/c244d0a2d53bf073f425202e.png"}, {"type": "image", "uri": "media/ur5fail/d395fa681965c470f97f04e2.png"}, {"type": "image", "uri": "media/ur5fail/23f45586b504766ea8cf68f7.png"}, {"type": "image", "uri": "media/ur5fail/b49fa3d41a3645b6255cdd2a.png"}, {"type": "image", "uri": "media/ur5fail/10e0caffc50ba4889bb96ed7.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0042/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 42, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/463f1948d8462670a8e72a4a.png"}, {"type": "image", "uri": "media/ur5fail/d8aaad67bf2d88c8114e80f1.png"}, {"type": "image", "uri": "media/ur5fail/c4d292eb42c497d220a00888.png"}, {"type": "image", "uri": "media/ur5fail/fdba99307c19024a2cee4497.png"}, {"type": "image", "uri": "media/ur5fail/e1d17066b00a318014cd9c76.png"}, {"type": "image", "uri": "media/ur5fail/588a2c7d0873e35c8ea6b8eb.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0045/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 45, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "stack the lemon and the peach on the strawberry", "subtask": "move grasped object on top of strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/3698d1fa12feafe685427420.png"}, {"type": "image", "uri": "media/ur5fail/333854d9225a1907e6d6fe68.png"}, {"type": "image", "uri": "media/ur5fail/e915904a9752dc3c2dfbfc66.png"}, {"type": "image", "uri": "media/ur5fail/b0f059d8034124667666c213.png"}, {"type": "image", "uri": "media/ur5fail/b6aa613fa55bd6f768fddac4.png"}, {"type": "image", "uri": "media/ur5fail/4f3ba8933ad7cffe2c80e9b6.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0047/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 47, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 347, "role": "error_localization"}}
{"state": {"task": "put the pink cube in the middle part of the drawer", "subtask": "grasp pink cube", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/440e0d84ddd260a6b6e7897b.png"}, {"type": "image", "uri": "media/ur5fail/861f6fc0d855b6aac7b54e44.png"}, {"type": "image", "uri": "media/ur5fail/a9f1366b935a7d647f43261d.png"}, {"type": "image", "uri": "media/ur5fail/e3ee7aac687880ac72987bb2.png"}, {"type": "image", "uri": "media/ur5fail/2a4ccc470e5f16f76d2421b6.png"}, {"type": "image", "uri": "media/ur5fail/b0cc0ef431e0a58944b41919.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0048/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+1/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 48, "taskvar": "real_put_item_in_drawer+1", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 345, "role": "error_localization"}}
{"state": {"task": "put the pink cube in the middle part of the drawer", "subtask": "open middle drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/440e0d84ddd260a6b6e7897b.png"}, {"type": "image", "uri": "media/ur5fail/861f6fc0d855b6aac7b54e44.png"}, {"type": "image", "uri": "media/ur5fail/a9f1366b935a7d647f43261d.png"}, {"type": "image", "uri": "media/ur5fail/e3ee7aac687880ac72987bb2.png"}, {"type": "image", "uri": "media/ur5fail/2a4ccc470e5f16f76d2421b6.png"}, {"type": "image", "uri": "media/ur5fail/b0cc0ef431e0a58944b41919.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0049/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+1/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 49, "taskvar": "real_put_item_in_drawer+1", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "put the ball, the frog toy, the can and the ball in the top part of the drawer", "subtask": "move top drawer handle to open position", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/5af13b1315f5576216ef7c2e.png"}, {"type": "image", "uri": "media/ur5fail/2957a5df1cc4d101de3277e3.png"}, {"type": "image", "uri": "media/ur5fail/3248863369d59fb2db2d1064.png"}, {"type": "image", "uri": "media/ur5fail/ce54c07a64162299e355a37f.png"}, {"type": "image", "uri": "media/ur5fail/bbaa3b960d03a4083d225856.png"}, {"type": "image", "uri": "media/ur5fail/f5c947f6069fde99f129d982.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0054/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 54, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 357, "role": "error_localization"}}
{"state": {"task": "put the ball, the frog toy, the can and the ball in the top part of the drawer", "subtask": "put the ball in the top part of the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ce54c07a64162299e355a37f.png"}, {"type": "image", "uri": "media/ur5fail/bbaa3b960d03a4083d225856.png"}, {"type": "image", "uri": "media/ur5fail/f5c947f6069fde99f129d982.png"}, {"type": "image", "uri": "media/ur5fail/0b2b9b721d004ac700d6219e.png"}, {"type": "image", "uri": "media/ur5fail/47ba64d4311f0a35aefd6ed3.png"}, {"type": "image", "uri": "media/ur5fail/d71fd3da02d52ea12f6f58b9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0055/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 55, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "put the frog toy, the ball, the can and the ball in the top part of the drawer", "subtask": "put the frog toy in the top part of the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ce54c07a64162299e355a37f.png"}, {"type": "image", "uri": "media/ur5fail/bbaa3b960d03a4083d225856.png"}, {"type": "image", "uri": "media/ur5fail/f5c947f6069fde99f129d982.png"}, {"type": "image", "uri": "media/ur5fail/64493ea0066bea5625a72999.png"}, {"type": "image", "uri": "media/ur5fail/c1e451812094f56810e45fdb.png"}, {"type": "image", "uri": "media/ur5fail/5d60e4d211d47d13095597c2.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0056/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 56, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 361, "role": "error_localization"}}
{"state": {"task": "put the ball, and the pink cube into the top part of the drawer", "subtask": "put the ball into the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/354efcd7a595da712ac03a62.png"}, {"type": "image", "uri": "media/ur5fail/8e31dbd9ed32a6cd94bed111.png"}, {"type": "image", "uri": "media/ur5fail/666637ad904b175114cfa0f7.png"}, {"type": "image", "uri": "media/ur5fail/8e6cc293729fdd2f3a21f3f8.png"}, {"type": "image", "uri": "media/ur5fail/00ce687b6c5fddee72f89381.png"}, {"type": "image", "uri": "media/ur5fail/1d16c3eba4d9d7d1a23af5ff.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0061/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 61, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "put the ball, and the pink cube into the top part of the drawer", "subtask": "grasp the ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/354efcd7a595da712ac03a62.png"}, {"type": "image", "uri": "media/ur5fail/8e31dbd9ed32a6cd94bed111.png"}, {"type": "image", "uri": "media/ur5fail/666637ad904b175114cfa0f7.png"}, {"type": "image", "uri": "media/ur5fail/50f0968134a4c492527e101f.png"}, {"type": "image", "uri": "media/ur5fail/10ee984e3c14e96355db4635.png"}, {"type": "image", "uri": "media/ur5fail/92a08cba8b3f23baf961c646.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0062/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 62, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 349, "role": "error_localization"}}
{"state": {"task": "put the pink cube and the frog toy into the top part of the drawer", "subtask": "put the pink cube in the top part of the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/43c909e2911aa447e0a6203f.png"}, {"type": "image", "uri": "media/ur5fail/94a49a84c27f44657950d0da.png"}, {"type": "image", "uri": "media/ur5fail/335be5aa8380f4c9bc6f6974.png"}, {"type": "image", "uri": "media/ur5fail/309f2d2c61bfe982e4c393ae.png"}, {"type": "image", "uri": "media/ur5fail/6dfec834cd26a58490d0a0a1.png"}, {"type": "image", "uri": "media/ur5fail/88cf1ace7e0e07b30e7c4105.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0063/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 63, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 356, "role": "error_localization"}}
{"state": {"task": "put the pink cube and the frog toy into the middle drawer", "subtask": "open middle drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/43c909e2911aa447e0a6203f.png"}, {"type": "image", "uri": "media/ur5fail/94a49a84c27f44657950d0da.png"}, {"type": "image", "uri": "media/ur5fail/335be5aa8380f4c9bc6f6974.png"}, {"type": "image", "uri": "media/ur5fail/309f2d2c61bfe982e4c393ae.png"}, {"type": "image", "uri": "media/ur5fail/6dfec834cd26a58490d0a0a1.png"}, {"type": "image", "uri": "media/ur5fail/88cf1ace7e0e07b30e7c4105.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0064/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 64, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 345, "role": "error_localization"}}
{"state": {"task": "put into the top drawer the frog toy, the ball, and pink cube", "subtask": "put into the top drawer the ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/b78e1dade212725d9df8e43f.png"}, {"type": "image", "uri": "media/ur5fail/beaf33286a831177023232ff.png"}, {"type": "image", "uri": "media/ur5fail/e89885e8d069f21a55f2c73b.png"}, {"type": "image", "uri": "media/ur5fail/3d7a35a47ea538c9f634c3db.png"}, {"type": "image", "uri": "media/ur5fail/5899ea49179bea304bd1de85.png"}, {"type": "image", "uri": "media/ur5fail/36f88e195bed7223a0e61946.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0068/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 68, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 352, "role": "error_localization"}}
{"state": {"task": "put into the middle drawer the frog toy, the ball, and pink cube", "subtask": "put into the middle drawer the frog toy", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/b78e1dade212725d9df8e43f.png"}, {"type": "image", "uri": "media/ur5fail/beaf33286a831177023232ff.png"}, {"type": "image", "uri": "media/ur5fail/e89885e8d069f21a55f2c73b.png"}, {"type": "image", "uri": "media/ur5fail/3d7a35a47ea538c9f634c3db.png"}, {"type": "image", "uri": "media/ur5fail/5899ea49179bea304bd1de85.png"}, {"type": "image", "uri": "media/ur5fail/36f88e195bed7223a0e61946.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0069/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 69, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "put into the bottom drawer the frog toy, the ball, and pink cube", "subtask": "put into the bottom drawer the frog toy", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/b78e1dade212725d9df8e43f.png"}, {"type": "image", "uri": "media/ur5fail/beaf33286a831177023232ff.png"}, {"type": "image", "uri": "media/ur5fail/e89885e8d069f21a55f2c73b.png"}, {"type": "image", "uri": "media/ur5fail/3d7a35a47ea538c9f634c3db.png"}, {"type": "image", "uri": "media/ur5fail/5899ea49179bea304bd1de85.png"}, {"type": "image", "uri": "media/ur5fail/36f88e195bed7223a0e61946.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0070/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 70, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 353, "role": "error_localization"}}
{"state": {"task": "put the ball in the top part of the drawer", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}, {"type": "image", "uri": "media/ur5fail/f8b669e7faa92bc28f1ed854.png"}, {"type": "image", "uri": "media/ur5fail/4ab698304e2d515e243251a1.png"}, {"type": "image", "uri": "media/ur5fail/82a16dd27111676943bb1030.png"}, {"type": "image", "uri": "media/ur5fail/e0c54f3c70e1e8cb996101f4.png"}, {"type": "image", "uri": "media/ur5fail/a491c9f4992bf015e441f42f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0071/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 71, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 343, "role": "error_localization"}}
{"state": {"task": "pick up the fruit between the cylinder and the cube", "subtask": "grasp lemon", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/f7ba6aa0c883551715c77b6d.png"}, {"type": "image", "uri": "media/ur5fail/c7c36a1941252b8c57beda98.png"}, {"type": "image", "uri": "media/ur5fail/d8528fe171162b2de38dd118.png"}, {"type": "image", "uri": "media/ur5fail/280a4237e07f60d8ff8264a3.png"}, {"type": "image", "uri": "media/ur5fail/d200ca8880bd48755a7576af.png"}, {"type": "image", "uri": "media/ur5fail/daa5d3c3360544589d609f3f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0073/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_box+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 73, "taskvar": "real_put_item_spatial_in_box+0", "episode_id": 0, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 343, "role": "error_localization"}}
{"state": {"task": "pick up the vegetable", "subtask": "grasp vegetable", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/1745db52a98aced3a79ab81b.png"}, {"type": "image", "uri": "media/ur5fail/00bd895937658778d80e366e.png"}, {"type": "image", "uri": "media/ur5fail/590d61b41d22d86512e24195.png"}, {"type": "image", "uri": "media/ur5fail/ce7fca1bc268dc059331c395.png"}, {"type": "image", "uri": "media/ur5fail/f397295792d19eeb40dfc5a6.png"}, {"type": "image", "uri": "media/ur5fail/af7ca8853b515e5e01404e92.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0074/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817412316", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 74, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817412316, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 337, "role": "error_localization"}}
{"state": {"task": "give me some pizza", "subtask": "grasp pizza", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/30eaecc29d9fef2fdf7cf88c.png"}, {"type": "image", "uri": "media/ur5fail/7e241a0a7b0cb22f9d37b6db.png"}, {"type": "image", "uri": "media/ur5fail/77c170d19d22916f1c6412a4.png"}, {"type": "image", "uri": "media/ur5fail/197204d71a1318eee15ae4b8.png"}, {"type": "image", "uri": "media/ur5fail/6032b5750045c7061f5b945a.png"}, {"type": "image", "uri": "media/ur5fail/af70c51fdf241383312ba777.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0076/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816477261", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 76, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816477261, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 337, "role": "error_localization"}}
{"state": {"task": "pick up the strawberry and place it in the blue plate", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/8c8ac4743d6ac066ceb1c36f.png"}, {"type": "image", "uri": "media/ur5fail/4a8f1e0d9f06ef465cbbc7c4.png"}, {"type": "image", "uri": "media/ur5fail/8b819083a3da55bb12b8a835.png"}, {"type": "image", "uri": "media/ur5fail/c654d4badd296298bd66833d.png"}, {"type": "image", "uri": "media/ur5fail/163adc660bb35dd3be6f3c9d.png"}, {"type": "image", "uri": "media/ur5fail/837ed2a5eae03e2092dd20a4.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0077/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816484059", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 77, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816484059, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "pick the pear and place it in the container", "subtask": "grasp pear", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/63301fad92e9e8cbbea783c3.png"}, {"type": "image", "uri": "media/ur5fail/aea00e2d23c46b64ac7ea662.png"}, {"type": "image", "uri": "media/ur5fail/b91e8f4e6ff9bf3db8b1a83c.png"}, {"type": "image", "uri": "media/ur5fail/864ad19a1a44505479230936.png"}, {"type": "image", "uri": "media/ur5fail/ca85fb5bbe842b0aba4581e3.png"}, {"type": "image", "uri": "media/ur5fail/10b9fc2bcdf63472804b366d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0080/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816542675", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 80, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816542675, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "pick the ball and place it in the box", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/3a5a74a967478def2803bd97.png"}, {"type": "image", "uri": "media/ur5fail/392946f7dd4736e5d837267e.png"}, {"type": "image", "uri": "media/ur5fail/c458b3278224fb25ebb30866.png"}, {"type": "image", "uri": "media/ur5fail/34705fd86374fa0c0a1833a9.png"}, {"type": "image", "uri": "media/ur5fail/72457308fec03975fa5fd79e.png"}, {"type": "image", "uri": "media/ur5fail/0f05ec355a1dae1e3c1a494b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0081/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816546187", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 81, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816546187, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "place in the box the donut", "subtask": "grasp donut", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ae77aab43a373759ee9c5599.png"}, {"type": "image", "uri": "media/ur5fail/5d916adae8dc9bd8be8024d6.png"}, {"type": "image", "uri": "media/ur5fail/a5aeddc031edb71762cfabfb.png"}, {"type": "image", "uri": "media/ur5fail/38dcd0e6aab390159ad1c277.png"}, {"type": "image", "uri": "media/ur5fail/1a686f8089d46bb574ed56b8.png"}, {"type": "image", "uri": "media/ur5fail/cedf649416d4792f3b266e27.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0085/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816572150", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 85, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816572150, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 341, "role": "error_localization"}}
{"state": {"task": "grab the bread", "subtask": "grasp bread", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/a5aa0fee63a2954e64f83a81.png"}, {"type": "image", "uri": "media/ur5fail/7aca91b52da15d8e05e9f060.png"}, {"type": "image", "uri": "media/ur5fail/ac2b8647caa7887829ea05fd.png"}, {"type": "image", "uri": "media/ur5fail/3fb27b5269c0db10407969fa.png"}, {"type": "image", "uri": "media/ur5fail/d7e8425c950a735204da7573.png"}, {"type": "image", "uri": "media/ur5fail/ca319868359898b9b0387792.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0089/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817368328", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 89, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817368328, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 336, "role": "error_localization"}}
{"state": {"task": "put the ball next to the carrot", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/458ccd670843141efd784cbd.png"}, {"type": "image", "uri": "media/ur5fail/ed75defdd6ceb5350129dd89.png"}, {"type": "image", "uri": "media/ur5fail/5f8aaa25f13311be07f8f197.png"}, {"type": "image", "uri": "media/ur5fail/8a5146e75c28feb7b76a1f13.png"}, {"type": "image", "uri": "media/ur5fail/125a62d2b87f4c6f687225fe.png"}, {"type": "image", "uri": "media/ur5fail/5bde157afd9bcc21cd619d7e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0091/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_sport+0/173817680775", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 91, "taskvar": "real_sport+0", "episode_id": 173817680775, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 340, "role": "error_localization"}}
{"state": {"task": "stack a black cup onto a red cup", "subtask": "grasp black cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/81633cb1b3536fc48d1501ee.png"}, {"type": "image", "uri": "media/ur5fail/06c3b29a69c09d25446e0229.png"}, {"type": "image", "uri": "media/ur5fail/c295e8c91be57af41942950c.png"}, {"type": "image", "uri": "media/ur5fail/028c93cd702b9ed2ff432a27.png"}, {"type": "image", "uri": "media/ur5fail/2263de9fc61e49c4ead1b9e6.png"}, {"type": "image", "uri": "media/ur5fail/baa0cf41a9fc70c92f56a335.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0093/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 93, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "stack black cup on orange cup", "subtask": "grasp black cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/c24d8e97cccbe22ab223bb25.png"}, {"type": "image", "uri": "media/ur5fail/9e75b87b143305195f656823.png"}, {"type": "image", "uri": "media/ur5fail/5c45fb2ccf71083ee6f06165.png"}, {"type": "image", "uri": "media/ur5fail/e93fc4e84288204d8373f8c2.png"}, {"type": "image", "uri": "media/ur5fail/06f0eed14e3bff6a16ff9040.png"}, {"type": "image", "uri": "media/ur5fail/80b6c36a7f1456cf4f001ceb.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0099/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_stack_cup+5/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 99, "taskvar": "real_stack_cup+5", "episode_id": 0, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 340, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box, then put the corn in the box", "subtask": "grasp corn", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/4c1ee0712f052cac143685ad.png"}, {"type": "image", "uri": "media/ur5fail/6787346e8ad2e8d90da97783.png"}, {"type": "image", "uri": "media/ur5fail/1b49c2ae66eb59c71ac5730b.png"}, {"type": "image", "uri": "media/ur5fail/5a402851e9708bfd1091cb75.png"}, {"type": "image", "uri": "media/ur5fail/6595ae26f243c85ebf2a2f37.png"}, {"type": "image", "uri": "media/ur5fail/9be3bc5cf8a56bca7f764edf.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0104/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/2", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 104, "taskvar": "real_put_food_in_box+0", "episode_id": 2, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 348, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box, then put the corn in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/5a402851e9708bfd1091cb75.png"}, {"type": "image", "uri": "media/ur5fail/6595ae26f243c85ebf2a2f37.png"}, {"type": "image", "uri": "media/ur5fail/9be3bc5cf8a56bca7f764edf.png"}, {"type": "image", "uri": "media/ur5fail/c1f966e1d9cd5f45714d9e17.png"}, {"type": "image", "uri": "media/ur5fail/85d7e6cecc8d7f10bd3f26ed.png"}, {"type": "image", "uri": "media/ur5fail/eb48dc4c059f44110cc997fa.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0105/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/2", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 105, "taskvar": "real_put_food_in_box+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box then put the banana in the box", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/21133fdf067010dcfc8dee38.png"}, {"type": "image", "uri": "media/ur5fail/e730ed4442068de5bd5ffaff.png"}, {"type": "image", "uri": "media/ur5fail/7737934e674f58110695df81.png"}, {"type": "image", "uri": "media/ur5fail/2170201391335c830c06c587.png"}, {"type": "image", "uri": "media/ur5fail/967d242912eb1de47507765c.png"}, {"type": "image", "uri": "media/ur5fail/997a2e84e5da9aa3dd81a748.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0107/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 107, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 346, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box then put the banana in the box", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/096e7fabaf984d3c8ef15c9f.png"}, {"type": "image", "uri": "media/ur5fail/42d41d0c7fa7fafa7177ff31.png"}, {"type": "image", "uri": "media/ur5fail/eb9f966c8119a8716c55dc47.png"}, {"type": "image", "uri": "media/ur5fail/d013101b9146b81f4a9a1667.png"}, {"type": "image", "uri": "media/ur5fail/bc35e990b01c83c43a39a5fd.png"}, {"type": "image", "uri": "media/ur5fail/cc4ad03ff36233b3254872ec.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0108/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 108, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 346, "role": "error_localization"}}
{"state": {"task": "pick the pink cup and put it onto the cyan cup", "subtask": "grasp pink cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/d8cd563714ee4dc92707a752.png"}, {"type": "image", "uri": "media/ur5fail/ed168a91fc834eb224b84f1d.png"}, {"type": "image", "uri": "media/ur5fail/46bf29ec1935e2dbe6dc3d6b.png"}, {"type": "image", "uri": "media/ur5fail/59b0e58f747668c010cac56a.png"}, {"type": "image", "uri": "media/ur5fail/42d10d70b33c8ff069677e01.png"}, {"type": "image", "uri": "media/ur5fail/59d65e104b9cb5f48c6482b5.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0111/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_stack_cups+2/10", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 111, "taskvar": "real_stack_cups+2", "episode_id": 10, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 345, "role": "error_localization"}}
{"state": {"task": "put the green fruit in the box", "subtask": "grasp green pear", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e5057a0d51fbdbd3ed4948c7.png"}, {"type": "image", "uri": "media/ur5fail/0ab870b13ff9c60649b6602f.png"}, {"type": "image", "uri": "media/ur5fail/ad270b30fb9464d4ab3037eb.png"}, {"type": "image", "uri": "media/ur5fail/0862f4ad24023fb2f675e5cd.png"}, {"type": "image", "uri": "media/ur5fail/bd36cfe877ace0d589546739.png"}, {"type": "image", "uri": "media/ur5fail/d7a58b8d7ab988568b368555.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0113/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 113, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 341, "role": "error_localization"}}
{"state": {"task": "put the fruits in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/8b59dbbea44ae020f107c2e5.png"}, {"type": "image", "uri": "media/ur5fail/87b888552c05c95c2dff927c.png"}, {"type": "image", "uri": "media/ur5fail/fb2ae8573c8b66ad4fc155b1.png"}, {"type": "image", "uri": "media/ur5fail/3181712a0228c1d988a523ad.png"}, {"type": "image", "uri": "media/ur5fail/621c2d1ac78d338765f10b4c.png"}, {"type": "image", "uri": "media/ur5fail/3f4d03eeefdd5007e69e0321.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0115/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/12", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 115, "taskvar": "real_put_food_in_box+0", "episode_id": 12, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "translation_target", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box with the strawberry", "subtask": "grasp tuna can", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/c912b9da87b713654190e29f.png"}, {"type": "image", "uri": "media/ur5fail/22fb2565f53d341ac3666fcd.png"}, {"type": "image", "uri": "media/ur5fail/dea2c3cd84185528dc5e7b98.png"}, {"type": "image", "uri": "media/ur5fail/049c1bab68a1672f379e1bd2.png"}, {"type": "image", "uri": "media/ur5fail/39df22c7d127d77df91b2bdf.png"}, {"type": "image", "uri": "media/ur5fail/c414b8d61aa51367efe3f4f1.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0117/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/12", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 117, "taskvar": "real_put_food_in_box+0", "episode_id": 12, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box with the strawberry", "subtask": "grasp tuna can", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/c912b9da87b713654190e29f.png"}, {"type": "image", "uri": "media/ur5fail/22fb2565f53d341ac3666fcd.png"}, {"type": "image", "uri": "media/ur5fail/dea2c3cd84185528dc5e7b98.png"}, {"type": "image", "uri": "media/ur5fail/bb18c321386820619bd331fc.png"}, {"type": "image", "uri": "media/ur5fail/a7b018f138f5e8830afa3b39.png"}, {"type": "image", "uri": "media/ur5fail/8e8031a8bb75965679423f8d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0118/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/12", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 118, "taskvar": "real_put_food_in_box+0", "episode_id": 12, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "put corn in the box with the tuna can and peach", "subtask": "grasp corn", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/a1b518f5308fb61dc4aaff08.png"}, {"type": "image", "uri": "media/ur5fail/b4662d4bd0b8146a826cab4f.png"}, {"type": "image", "uri": "media/ur5fail/085d780be994d28aa55d63e3.png"}, {"type": "image", "uri": "media/ur5fail/a001f41b7329f074400e1cad.png"}, {"type": "image", "uri": "media/ur5fail/ae9efc31b09e97e57f58c267.png"}, {"type": "image", "uri": "media/ur5fail/c03a46016863c5d847c097f3.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0119/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/13", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 119, "taskvar": "real_put_food_in_box+0", "episode_id": 13, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "my house is broken, please fix it", "subtask": "grasp hammer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/4f6496608f3b9e7093297875.png"}, {"type": "image", "uri": "media/ur5fail/2fa7ac7648dcb09b0aaadd47.png"}, {"type": "image", "uri": "media/ur5fail/52a3a7c67b8d2c3a81cdd326.png"}, {"type": "image", "uri": "media/ur5fail/4f6496608f3b9e7093297875.png"}, {"type": "image", "uri": "media/ur5fail/2fa7ac7648dcb09b0aaadd47.png"}, {"type": "image", "uri": "media/ur5fail/52a3a7c67b8d2c3a81cdd326.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0120/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_pick_tool_gondola+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 120, "taskvar": "real_pick_tool_gondola+0", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 341, "role": "error_localization"}}
{"state": {"task": "clean the table by putting things in the small box", "subtask": "move grasped object to small container", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_state_or_placement", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/eeaaf69cdcd33cba3e056e62.png"}, {"type": "image", "uri": "media/ur5fail/832d3784e19beef74acd473e.png"}, {"type": "image", "uri": "media/ur5fail/8f1ae6cf12b78a3e480419e0.png"}, {"type": "image", "uri": "media/ur5fail/6e5913725fb1f44939a26f89.png"}, {"type": "image", "uri": "media/ur5fail/be3791c13aa3a2e4f93ba6ed.png"}, {"type": "image", "uri": "media/ur5fail/8591a1732e75b4b83ecb271b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0123/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 123, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "slip", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 347, "role": "error_localization"}}
{"state": {"task": "put everything in the big box", "subtask": "grasp fruit", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/2368503e919d3001e52b1354.png"}, {"type": "image", "uri": "media/ur5fail/8e53800ca67452243e8188e5.png"}, {"type": "image", "uri": "media/ur5fail/8869d9904265d82d251997d0.png"}, {"type": "image", "uri": "media/ur5fail/4ffeb2fbc92eb6de336550b7.png"}, {"type": "image", "uri": "media/ur5fail/037da0bef756d0df08b55ec5.png"}, {"type": "image", "uri": "media/ur5fail/07fe3b58f2ab85993bfe4a11.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0124/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/1", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 124, "taskvar": "real_put_carrot_in_container+0", "episode_id": 1, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 339, "role": "error_localization"}}
{"state": {"task": "put everything in the big box", "subtask": "grasp fruit", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}, {"type": "image", "uri": "media/ur5fail/d8a71acbfa3bafd2b32d8c8c.png"}, {"type": "image", "uri": "media/ur5fail/f1207aac490261d7754d08d4.png"}, {"type": "image", "uri": "media/ur5fail/f7e289349d199be148c1fdbb.png"}, {"type": "image", "uri": "media/ur5fail/fb760c1210fa63750a761b9c.png"}, {"type": "image", "uri": "media/ur5fail/9dc053cf22541d79f7dd2648.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0125/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 125, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 339, "role": "error_localization"}}
{"state": {"task": "clean the things on the table out of the table", "subtask": "grasp white plate", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/5af5193de747ee3e8789b355.png"}, {"type": "image", "uri": "media/ur5fail/2a63444c2fd1219ccb39f44c.png"}, {"type": "image", "uri": "media/ur5fail/6c31269155ea4856245e6ef0.png"}, {"type": "image", "uri": "media/ur5fail/993a535ff8cee0329ba6ce82.png"}, {"type": "image", "uri": "media/ur5fail/e59855253a9adf9a185c7b08.png"}, {"type": "image", "uri": "media/ur5fail/ce3f174a3324600176291949.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0127/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_carrot_in_plate+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 127, "taskvar": "real_put_carrot_in_plate+0", "episode_id": 0, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "stack the pink cup onto the black cup then the black cup onto the cyan cup, finally stack the cyan cup onto the yellow cup", "subtask": "grasp pink cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e4e68536c5d77ddb2f0cb3fd.png"}, {"type": "image", "uri": "media/ur5fail/894e112a24b9eb34c100bd17.png"}, {"type": "image", "uri": "media/ur5fail/dcb4ab8b1d35b4a3c685eb00.png"}, {"type": "image", "uri": "media/ur5fail/908bba46dd5f97b1e6483706.png"}, {"type": "image", "uri": "media/ur5fail/36cbd115b11d8caff7f7d75e.png"}, {"type": "image", "uri": "media/ur5fail/6a4b540082ad4fa99447a590.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0128/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_stack_cup+3/5", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 128, "taskvar": "real_stack_cup+3", "episode_id": 5, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/db912aa59345d929bbd360c8.png"}, {"type": "image", "uri": "media/ur5fail/58ace0d98d14e15c58f29c00.png"}, {"type": "image", "uri": "media/ur5fail/876666ea98e6c485dda1fc0b.png"}, {"type": "image", "uri": "media/ur5fail/deff1af6cd9bb45e4629350c.png"}, {"type": "image", "uri": "media/ur5fail/f5cb388c8b8c209b171ede93.png"}, {"type": "image", "uri": "media/ur5fail/1ed7f4a50c2c350eb7e8cf94.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0129/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/2", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 129, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 2, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/d32b55e6429bafd98990271a.png"}, {"type": "image", "uri": "media/ur5fail/049bd6cfdb13be69a1b2dbba.png"}, {"type": "image", "uri": "media/ur5fail/4f7869f3274b449e0c26f8c8.png"}, {"type": "image", "uri": "media/ur5fail/5de64c7e5f3d62a2e777fd8e.png"}, {"type": "image", "uri": "media/ur5fail/1dd95f753835f5c5aad10c57.png"}, {"type": "image", "uri": "media/ur5fail/98dbed86a1523ae2d6ac5a10.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0130/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/3", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 130, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 3, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "put the pink plate on top of the navy plate, and move the healthiest food to the pink plate", "subtask": "grasp pink plate", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "imprecise_grasp_or_push", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/aa26ac7fe092ff0e6d5b9978.png"}, {"type": "image", "uri": "media/ur5fail/9de7667295a8a01c76bf0a34.png"}, {"type": "image", "uri": "media/ur5fail/0795bd2a268cf77a48ccf76c.png"}, {"type": "image", "uri": "media/ur5fail/79148c63aeaa0b1326635ed5.png"}, {"type": "image", "uri": "media/ur5fail/81feb4854ee70b7bcb53a456.png"}, {"type": "image", "uri": "media/ur5fail/8221e636992774bb6433aa89.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0131/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_healthiest_banana_in_plate+1/7", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 131, "taskvar": "real_healthiest_banana_in_plate+1", "episode_id": 7, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 355, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp healthiest food", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "wrong_object", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/2cc4c67d22468ce90cd65273.png"}, {"type": "image", "uri": "media/ur5fail/02a24eba277a34207592757a.png"}, {"type": "image", "uri": "media/ur5fail/3feafbe07ce3834b8c26418a.png"}, {"type": "image", "uri": "media/ur5fail/caeb7759f2f0a0aa7a80a73c.png"}, {"type": "image", "uri": "media/ur5fail/dabb349694fe01913ebf8036.png"}, {"type": "image", "uri": "media/ur5fail/39a62b0fcf7b854244d0612e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0133/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_healthiest_in_plate+0/0", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 133, "taskvar": "real_healthiest_in_plate+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "put the scissors and the carrot in the container", "subtask": "grasp scissor", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "choice", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. This subtask is known to have FAILED. Which category best describes the failure?", "criteria": {"no_gripper_close": "No gripper close: the gripper is correctly positioned to grasp the object, but it did not close its jaws.", "imprecise_grasp_or_push": "Imprecise grasping/pushing: the gripper attempted to grasp or push the correct object by moving toward it and closing its jaws, but missed due to inaccurate positioning.", "wrong_state_or_placement": "Wrong object state or placement: the gripper tried to move, release, or change the state of the correct object, but the final placement or state is wrong.", "wrong_object": "Wrong object manipulated: the gripper manipulated the wrong object.", "no_progress": "No progress: neither the scene state nor the robot's configuration changed in any meaningful way."}, "label": "no_gripper_close", "src": "ur5fail_exec_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/851d71871ac2f52df82e11fa.png"}, {"type": "image", "uri": "media/ur5fail/18fe98f12d2ffe7956e3f94a.png"}, {"type": "image", "uri": "media/ur5fail/7103329de79eb58d722a45d8.png"}, {"type": "image", "uri": "media/ur5fail/453f0e02da46926e09b418f0.png"}, {"type": "image", "uri": "media/ur5fail/3f6b315328597534c78bd4e8.png"}, {"type": "image", "uri": "media/ur5fail/4d7944217379d911c883f88c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0139/ur5fail_exec_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_exec_failure_mode", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 139, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 343, "role": "error_localization"}}
{"state": {"task": "pick up the fruit, and lift it up, then put it back on the table, then put it in the on the floor", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ad4501b9673211a57736fdcf.png"}, {"type": "image", "uri": "media/ur5fail/5ac9c43e9ffe6922926f8495.png"}, {"type": "image", "uri": "media/ur5fail/41a6fd56ee1c087689219044.png"}, {"type": "image", "uri": "media/ur5fail/d10db5453792507257c1e727.png"}, {"type": "image", "uri": "media/ur5fail/7c704973d6594c0403eac985.png"}, {"type": "image", "uri": "media/ur5fail/b5f4b7773eb634d1928b22a5.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0000/ur5fail_exec_success", "group_id": "ur5fail/real_pick_tool+0/173817592039", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 0, "taskvar": "real_pick_tool+0", "episode_id": 173817592039, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 287, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box, then move out the tuna can and put it back on table, and put the tuna can on the floor", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/2b313fa95f6d493bc05b8185.png"}, {"type": "image", "uri": "media/ur5fail/d3b0be87528ac433ae5c31e2.png"}, {"type": "image", "uri": "media/ur5fail/e6f5e3aed9ca9587f282929f.png"}, {"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0001/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 1, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 300, "role": "outcome"}}
{"state": {"task": "pick up the ball", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/f349279bbeb2818226a25b3f.png"}, {"type": "image", "uri": "media/ur5fail/542104e9c1edbe8526c95876.png"}, {"type": "image", "uri": "media/ur5fail/d20ad84fb548022298500623.png"}, {"type": "image", "uri": "media/ur5fail/59762efe252de42dc815e8f6.png"}, {"type": "image", "uri": "media/ur5fail/61e702940222756c349615a6.png"}, {"type": "image", "uri": "media/ur5fail/6ae940908fac4de2519127fa.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0002/ur5fail_exec_success", "group_id": "ur5fail/real_pick_tool+0/173817611377", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 2, "taskvar": "real_pick_tool+0", "episode_id": 173817611377, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 265, "role": "outcome"}}
{"state": {"task": "grasp the lightest object", "subtask": "grasp the lightest object", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/f349279bbeb2818226a25b3f.png"}, {"type": "image", "uri": "media/ur5fail/542104e9c1edbe8526c95876.png"}, {"type": "image", "uri": "media/ur5fail/d20ad84fb548022298500623.png"}, {"type": "image", "uri": "media/ur5fail/59762efe252de42dc815e8f6.png"}, {"type": "image", "uri": "media/ur5fail/61e702940222756c349615a6.png"}, {"type": "image", "uri": "media/ur5fail/6ae940908fac4de2519127fa.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0003/ur5fail_exec_success", "group_id": "ur5fail/real_pick_tool+0/173817611377", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 3, "taskvar": "real_pick_tool+0", "episode_id": 173817611377, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "grasp the heaviest object", "subtask": "grasp hammer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/f349279bbeb2818226a25b3f.png"}, {"type": "image", "uri": "media/ur5fail/542104e9c1edbe8526c95876.png"}, {"type": "image", "uri": "media/ur5fail/d20ad84fb548022298500623.png"}, {"type": "image", "uri": "media/ur5fail/59762efe252de42dc815e8f6.png"}, {"type": "image", "uri": "media/ur5fail/61e702940222756c349615a6.png"}, {"type": "image", "uri": "media/ur5fail/6ae940908fac4de2519127fa.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0004/ur5fail_exec_success", "group_id": "ur5fail/real_pick_tool+0/173817611377", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 4, "taskvar": "real_pick_tool+0", "episode_id": 173817611377, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 267, "role": "outcome"}}
{"state": {"task": "pick up the ball", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ad4501b9673211a57736fdcf.png"}, {"type": "image", "uri": "media/ur5fail/5ac9c43e9ffe6922926f8495.png"}, {"type": "image", "uri": "media/ur5fail/41a6fd56ee1c087689219044.png"}, {"type": "image", "uri": "media/ur5fail/d10db5453792507257c1e727.png"}, {"type": "image", "uri": "media/ur5fail/7c704973d6594c0403eac985.png"}, {"type": "image", "uri": "media/ur5fail/b5f4b7773eb634d1928b22a5.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0005/ur5fail_exec_success", "group_id": "ur5fail/real_pick_tool+0/173817592039", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 5, "taskvar": "real_pick_tool+0", "episode_id": 173817592039, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 265, "role": "outcome"}}
{"state": {"task": "pick up the fruit, and lift it up, then put it back on the table", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ad4501b9673211a57736fdcf.png"}, {"type": "image", "uri": "media/ur5fail/5ac9c43e9ffe6922926f8495.png"}, {"type": "image", "uri": "media/ur5fail/41a6fd56ee1c087689219044.png"}, {"type": "image", "uri": "media/ur5fail/d10db5453792507257c1e727.png"}, {"type": "image", "uri": "media/ur5fail/7c704973d6594c0403eac985.png"}, {"type": "image", "uri": "media/ur5fail/b5f4b7773eb634d1928b22a5.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0006/ur5fail_exec_success", "group_id": "ur5fail/real_pick_tool+0/173817592039", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 6, "taskvar": "real_pick_tool+0", "episode_id": 173817592039, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 278, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "subtask": "grasp tuna can", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}, {"type": "image", "uri": "media/ur5fail/b14edf6d62c5cf2ed5488855.png"}, {"type": "image", "uri": "media/ur5fail/aad32040564cbe94530b39b8.png"}, {"type": "image", "uri": "media/ur5fail/3bc30872fe8dcd494c0c4859.png"}, {"type": "image", "uri": "media/ur5fail/3258013a066c2f954b29e48d.png"}, {"type": "image", "uri": "media/ur5fail/acea5dc1d3dc6bcac0e38461.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0007/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 7, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 276, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}, {"type": "image", "uri": "media/ur5fail/2b313fa95f6d493bc05b8185.png"}, {"type": "image", "uri": "media/ur5fail/d3b0be87528ac433ae5c31e2.png"}, {"type": "image", "uri": "media/ur5fail/e6f5e3aed9ca9587f282929f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0008/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 8, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 278, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/2b313fa95f6d493bc05b8185.png"}, {"type": "image", "uri": "media/ur5fail/d3b0be87528ac433ae5c31e2.png"}, {"type": "image", "uri": "media/ur5fail/e6f5e3aed9ca9587f282929f.png"}, {"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0009/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 9, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 278, "role": "outcome"}}
{"state": {"task": "put the tuna can next to but not inside box then put the corn in the box", "subtask": "move grasped object next to but not inside box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/2b313fa95f6d493bc05b8185.png"}, {"type": "image", "uri": "media/ur5fail/d3b0be87528ac433ae5c31e2.png"}, {"type": "image", "uri": "media/ur5fail/e6f5e3aed9ca9587f282929f.png"}, {"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0010/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 10, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 285, "role": "outcome"}}
{"state": {"task": "put the tuna can to the pizza slice then put the corn in the box", "subtask": "move grasped object to the pizza slice", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}, {"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}, {"type": "image", "uri": "media/ur5fail/b14edf6d62c5cf2ed5488855.png"}, {"type": "image", "uri": "media/ur5fail/aad32040564cbe94530b39b8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0011/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 11, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 281, "role": "outcome"}}
{"state": {"task": "put the tuna can next to the strawberry then the strawberry in the box, then the donut in the box", "subtask": "move grasped object next to strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}, {"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0012/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 12, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 287, "role": "outcome"}}
{"state": {"task": "remove any object from the box and put it on the table", "subtask": "put tuna can on table", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}, {"type": "image", "uri": "media/ur5fail/3bc30872fe8dcd494c0c4859.png"}, {"type": "image", "uri": "media/ur5fail/3258013a066c2f954b29e48d.png"}, {"type": "image", "uri": "media/ur5fail/acea5dc1d3dc6bcac0e38461.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0013/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 13, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 275, "role": "outcome"}}
{"state": {"task": "put all the fruits into the box", "subtask": "put strawberry into the box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/3bc30872fe8dcd494c0c4859.png"}, {"type": "image", "uri": "media/ur5fail/3258013a066c2f954b29e48d.png"}, {"type": "image", "uri": "media/ur5fail/acea5dc1d3dc6bcac0e38461.png"}, {"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0014/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 14, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the pizza slice in the box", "subtask": "grasp pizza slice", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}, {"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0015/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 15, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 277, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the pizza slice in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}, {"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0016/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 16, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 279, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the corn cob between the strawberry and the donut", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}, {"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}, {"type": "image", "uri": "media/ur5fail/b14edf6d62c5cf2ed5488855.png"}, {"type": "image", "uri": "media/ur5fail/aad32040564cbe94530b39b8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0017/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 17, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 283, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "subtask": "grasp corn", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}, {"type": "image", "uri": "media/ur5fail/70c8c2e550dbfcba2a12a170.png"}, {"type": "image", "uri": "media/ur5fail/4bef3431062e072baa30f95c.png"}, {"type": "image", "uri": "media/ur5fail/f5c0212f6f9862ecf7f6860a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0018/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 18, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 275, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the black grapes in the box", "subtask": "grasp black grapes", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}, {"type": "image", "uri": "media/ur5fail/70c8c2e550dbfcba2a12a170.png"}, {"type": "image", "uri": "media/ur5fail/4bef3431062e072baa30f95c.png"}, {"type": "image", "uri": "media/ur5fail/f5c0212f6f9862ecf7f6860a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0019/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 19, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 277, "role": "outcome"}}
{"state": {"task": "grasp the corn cob, lift it up, and place it in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}, {"type": "image", "uri": "media/ur5fail/8855802adba6e4860c5317c5.png"}, {"type": "image", "uri": "media/ur5fail/f7d3d3e2939db1f59637fe92.png"}, {"type": "image", "uri": "media/ur5fail/5bd927b9bbf28508588d1c49.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0020/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 20, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 279, "role": "outcome"}}
{"state": {"task": "grasp the corn cob, lift it up, and place it in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/e8e8bf515f366e17bd672938.png"}, {"type": "image", "uri": "media/ur5fail/1d21d267202f96a93a2bda9d.png"}, {"type": "image", "uri": "media/ur5fail/c73beaf3bed72bcc4e3e1070.png"}, {"type": "image", "uri": "media/ur5fail/70c8c2e550dbfcba2a12a170.png"}, {"type": "image", "uri": "media/ur5fail/4bef3431062e072baa30f95c.png"}, {"type": "image", "uri": "media/ur5fail/f5c0212f6f9862ecf7f6860a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0021/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 21, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 279, "role": "outcome"}}
{"state": {"task": "grasp the tuna can, lift it up, and place it in the box, grasp the corn cob, lift it up, and place it in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}, {"type": "image", "uri": "media/ur5fail/8dc374669648ad41c41951da.png"}, {"type": "image", "uri": "media/ur5fail/4124180cb5561b6b86fb2e87.png"}, {"type": "image", "uri": "media/ur5fail/20baac838951a0e45b795c02.png"}, {"type": "image", "uri": "media/ur5fail/9765b437be0e28098fe0bf35.png"}, {"type": "image", "uri": "media/ur5fail/92c1be969bd5e31035652f5c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0022/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 22, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 295, "role": "outcome"}}
{"state": {"task": "grasp the tuna can, lift it up, and place it in the box, grasp the corn cob, lift it up, and place it in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}, {"type": "image", "uri": "media/ur5fail/60d16780c9068bf3e9292bc5.png"}, {"type": "image", "uri": "media/ur5fail/22dce9957f0f5bfc42486227.png"}, {"type": "image", "uri": "media/ur5fail/177a01339f884083162e5454.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0023/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 23, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 296, "role": "outcome"}}
{"state": {"task": "grasp the tuna can, lift it up, and place it in the box, grasp the corn cob, lift it up, and place it in the box, then put the corn cob back to the table", "subtask": "move grasped object down on the table", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}, {"type": "image", "uri": "media/ur5fail/60d16780c9068bf3e9292bc5.png"}, {"type": "image", "uri": "media/ur5fail/22dce9957f0f5bfc42486227.png"}, {"type": "image", "uri": "media/ur5fail/177a01339f884083162e5454.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0024/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 24, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 308, "role": "outcome"}}
{"state": {"task": "grasp the tuna can, lift it up, and place it in the box, grasp the corn cob, lift it up, and place it in the box, then put the corn cob back to the table", "subtask": "release", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/2b313fa95f6d493bc05b8185.png"}, {"type": "image", "uri": "media/ur5fail/d3b0be87528ac433ae5c31e2.png"}, {"type": "image", "uri": "media/ur5fail/e6f5e3aed9ca9587f282929f.png"}, {"type": "image", "uri": "media/ur5fail/6adb1d4b1db7ec0f814ae684.png"}, {"type": "image", "uri": "media/ur5fail/dffaf5490d907afbf144b75b.png"}, {"type": "image", "uri": "media/ur5fail/1712e11d42644ccfa9933029.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0025/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 25, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 301, "role": "outcome"}}
{"state": {"task": "grasp the tuna can, lift it up, and place it in the box, grasp the corn cob, lift it up, and place it in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}, {"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0026/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 26, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "no_progress", "failure_mode_raw": "no progress", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 296, "role": "outcome"}}
{"state": {"task": "grasp the donut, and place it in the box, grasp the corn cob, lift it up, and place it in the box", "subtask": "put the donut in the box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}, {"type": "image", "uri": "media/ur5fail/ef2ec0de6516a44aa5f27e65.png"}, {"type": "image", "uri": "media/ur5fail/362d402f9cc6a15495340d2f.png"}, {"type": "image", "uri": "media/ur5fail/2112e0fc2457d0caa90f1fb7.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0027/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 27, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 293, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box then put the donut in the box", "subtask": "grasp donut", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/56206ecfe1657377c4fe979a.png"}, {"type": "image", "uri": "media/ur5fail/0caec930fea30714f61b6135.png"}, {"type": "image", "uri": "media/ur5fail/d56efaec3b2fdae965bea502.png"}, {"type": "image", "uri": "media/ur5fail/c618787c6df83bd140f0b783.png"}, {"type": "image", "uri": "media/ur5fail/592dbd7a46cc410b1ae6119a.png"}, {"type": "image", "uri": "media/ur5fail/4c391536923d24ac64641d8b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0028/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/1", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 28, "taskvar": "real_put_food_in_box+0", "episode_id": 1, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 277, "role": "outcome"}}
{"state": {"task": "put the banana then the grapes in the box", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}, {"type": "image", "uri": "media/ur5fail/9641470ccc34dc3a51bebcd2.png"}, {"type": "image", "uri": "media/ur5fail/5019194395c260a6640fa98a.png"}, {"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0029/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 29, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the banana then the grapes in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}, {"type": "image", "uri": "media/ur5fail/d2ef187a2220dacb11a37992.png"}, {"type": "image", "uri": "media/ur5fail/97cee9d9fee6cfa594faf221.png"}, {"type": "image", "uri": "media/ur5fail/e1a2f706286eca5383b52bf7.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0030/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 30, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the banana in the box then put it back on the table", "subtask": "move grasped object down on the table", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d2ef187a2220dacb11a37992.png"}, {"type": "image", "uri": "media/ur5fail/97cee9d9fee6cfa594faf221.png"}, {"type": "image", "uri": "media/ur5fail/e1a2f706286eca5383b52bf7.png"}, {"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0031/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 31, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 279, "role": "outcome"}}
{"state": {"task": "put the banana in the box then put it back on the table", "subtask": "move grasped object down on the table", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d2ef187a2220dacb11a37992.png"}, {"type": "image", "uri": "media/ur5fail/97cee9d9fee6cfa594faf221.png"}, {"type": "image", "uri": "media/ur5fail/e1a2f706286eca5383b52bf7.png"}, {"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}, {"type": "image", "uri": "media/ur5fail/9641470ccc34dc3a51bebcd2.png"}, {"type": "image", "uri": "media/ur5fail/5019194395c260a6640fa98a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0032/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 32, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 279, "role": "outcome"}}
{"state": {"task": "put the grapes in the box then the banana", "subtask": "grasp grapes", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}, {"type": "image", "uri": "media/ur5fail/9641470ccc34dc3a51bebcd2.png"}, {"type": "image", "uri": "media/ur5fail/5019194395c260a6640fa98a.png"}, {"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0033/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 33, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the strawberry in the box then the pear", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}, {"type": "image", "uri": "media/ur5fail/9641470ccc34dc3a51bebcd2.png"}, {"type": "image", "uri": "media/ur5fail/5019194395c260a6640fa98a.png"}, {"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0034/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 34, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the banana then the grapes in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}, {"type": "image", "uri": "media/ur5fail/1ea98f9ac267d7c5bb96e975.png"}, {"type": "image", "uri": "media/ur5fail/9d5aa6e529c73e1cd6380f3e.png"}, {"type": "image", "uri": "media/ur5fail/a734626ff8b4b891b9408f5e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0035/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 35, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 273, "role": "outcome"}}
{"state": {"task": "put the grapes then the banana in the box, then put them back on the table", "subtask": "move grasped object down to table", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/1ea98f9ac267d7c5bb96e975.png"}, {"type": "image", "uri": "media/ur5fail/9d5aa6e529c73e1cd6380f3e.png"}, {"type": "image", "uri": "media/ur5fail/a734626ff8b4b891b9408f5e.png"}, {"type": "image", "uri": "media/ur5fail/021c378b17d119fc8c5d1220.png"}, {"type": "image", "uri": "media/ur5fail/d25255b5ae92f65581f65ff6.png"}, {"type": "image", "uri": "media/ur5fail/5d0812578f1217db31b3eb6a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0036/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 36, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 282, "role": "outcome"}}
{"state": {"task": "put the banana then the grapes in the box, then put them back on the table", "subtask": "release", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/faa95603cf9dde8e979a631f.png"}, {"type": "image", "uri": "media/ur5fail/128b0efb96293dfa79b0297f.png"}, {"type": "image", "uri": "media/ur5fail/9c91b58f97136cf6a3de32c6.png"}, {"type": "image", "uri": "media/ur5fail/9d1ecd236442311b1e345803.png"}, {"type": "image", "uri": "media/ur5fail/9bc0041695b48ac00f3d1ad4.png"}, {"type": "image", "uri": "media/ur5fail/f3e49635bee479b8c920552f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0037/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 37, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 276, "role": "outcome"}}
{"state": {"task": "put the banana between yellow peach and strawberry, then put the black grapes in the box", "subtask": "move grasped object to between yellow peach and strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/faa95603cf9dde8e979a631f.png"}, {"type": "image", "uri": "media/ur5fail/128b0efb96293dfa79b0297f.png"}, {"type": "image", "uri": "media/ur5fail/9c91b58f97136cf6a3de32c6.png"}, {"type": "image", "uri": "media/ur5fail/9d1ecd236442311b1e345803.png"}, {"type": "image", "uri": "media/ur5fail/9bc0041695b48ac00f3d1ad4.png"}, {"type": "image", "uri": "media/ur5fail/f3e49635bee479b8c920552f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0038/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 38, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 285, "role": "outcome"}}
{"state": {"task": "put the banana then the grapes in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/78956582976ac25f24688bd5.png"}, {"type": "image", "uri": "media/ur5fail/6935c0ca98c3b7c6ece18b43.png"}, {"type": "image", "uri": "media/ur5fail/e28ee351118bf8d9f18522c8.png"}, {"type": "image", "uri": "media/ur5fail/e5f9ca797dcc4f21016440d6.png"}, {"type": "image", "uri": "media/ur5fail/a9cc6e2dde9f14c146327605.png"}, {"type": "image", "uri": "media/ur5fail/8faca1d299e4c9584c20be95.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0039/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 39, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 273, "role": "outcome"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "grasp lemon", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}, {"type": "image", "uri": "media/ur5fail/c244d0a2d53bf073f425202e.png"}, {"type": "image", "uri": "media/ur5fail/d395fa681965c470f97f04e2.png"}, {"type": "image", "uri": "media/ur5fail/8e2760d2c3383c0c8ec46633.png"}, {"type": "image", "uri": "media/ur5fail/258ec3d354e3df345d6e68de.png"}, {"type": "image", "uri": "media/ur5fail/2e0fcc2ef63b9ec96ed7bca0.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0040/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 40, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "grasp lemon", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}, {"type": "image", "uri": "media/ur5fail/c244d0a2d53bf073f425202e.png"}, {"type": "image", "uri": "media/ur5fail/d395fa681965c470f97f04e2.png"}, {"type": "image", "uri": "media/ur5fail/3698d1fa12feafe685427420.png"}, {"type": "image", "uri": "media/ur5fail/333854d9225a1907e6d6fe68.png"}, {"type": "image", "uri": "media/ur5fail/e915904a9752dc3c2dfbfc66.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0041/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 41, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "grasp lemon", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}, {"type": "image", "uri": "media/ur5fail/c244d0a2d53bf073f425202e.png"}, {"type": "image", "uri": "media/ur5fail/d395fa681965c470f97f04e2.png"}, {"type": "image", "uri": "media/ur5fail/23f45586b504766ea8cf68f7.png"}, {"type": "image", "uri": "media/ur5fail/b49fa3d41a3645b6255cdd2a.png"}, {"type": "image", "uri": "media/ur5fail/10e0caffc50ba4889bb96ed7.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0042/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 42, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/8e2760d2c3383c0c8ec46633.png"}, {"type": "image", "uri": "media/ur5fail/258ec3d354e3df345d6e68de.png"}, {"type": "image", "uri": "media/ur5fail/2e0fcc2ef63b9ec96ed7bca0.png"}, {"type": "image", "uri": "media/ur5fail/b0f059d8034124667666c213.png"}, {"type": "image", "uri": "media/ur5fail/b6aa613fa55bd6f768fddac4.png"}, {"type": "image", "uri": "media/ur5fail/4f3ba8933ad7cffe2c80e9b6.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0043/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 43, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 273, "role": "outcome"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/463f1948d8462670a8e72a4a.png"}, {"type": "image", "uri": "media/ur5fail/d8aaad67bf2d88c8114e80f1.png"}, {"type": "image", "uri": "media/ur5fail/c4d292eb42c497d220a00888.png"}, {"type": "image", "uri": "media/ur5fail/013cb0c32d07a4792c97c5f1.png"}, {"type": "image", "uri": "media/ur5fail/35193bc34a50933f524d324b.png"}, {"type": "image", "uri": "media/ur5fail/06a478edf0f2e67b5aa2503b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0044/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 44, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/463f1948d8462670a8e72a4a.png"}, {"type": "image", "uri": "media/ur5fail/d8aaad67bf2d88c8114e80f1.png"}, {"type": "image", "uri": "media/ur5fail/c4d292eb42c497d220a00888.png"}, {"type": "image", "uri": "media/ur5fail/fdba99307c19024a2cee4497.png"}, {"type": "image", "uri": "media/ur5fail/e1d17066b00a318014cd9c76.png"}, {"type": "image", "uri": "media/ur5fail/588a2c7d0873e35c8ea6b8eb.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0045/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 45, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the lemon then the peach in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/3698d1fa12feafe685427420.png"}, {"type": "image", "uri": "media/ur5fail/333854d9225a1907e6d6fe68.png"}, {"type": "image", "uri": "media/ur5fail/e915904a9752dc3c2dfbfc66.png"}, {"type": "image", "uri": "media/ur5fail/b0f059d8034124667666c213.png"}, {"type": "image", "uri": "media/ur5fail/b6aa613fa55bd6f768fddac4.png"}, {"type": "image", "uri": "media/ur5fail/4f3ba8933ad7cffe2c80e9b6.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0046/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 46, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 273, "role": "outcome"}}
{"state": {"task": "stack the lemon and the peach on the strawberry", "subtask": "move grasped object on top of strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/3698d1fa12feafe685427420.png"}, {"type": "image", "uri": "media/ur5fail/333854d9225a1907e6d6fe68.png"}, {"type": "image", "uri": "media/ur5fail/e915904a9752dc3c2dfbfc66.png"}, {"type": "image", "uri": "media/ur5fail/b0f059d8034124667666c213.png"}, {"type": "image", "uri": "media/ur5fail/b6aa613fa55bd6f768fddac4.png"}, {"type": "image", "uri": "media/ur5fail/4f3ba8933ad7cffe2c80e9b6.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0047/ur5fail_exec_success", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 47, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 275, "role": "outcome"}}
{"state": {"task": "put the pink cube in the middle part of the drawer", "subtask": "grasp pink cube", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/440e0d84ddd260a6b6e7897b.png"}, {"type": "image", "uri": "media/ur5fail/861f6fc0d855b6aac7b54e44.png"}, {"type": "image", "uri": "media/ur5fail/a9f1366b935a7d647f43261d.png"}, {"type": "image", "uri": "media/ur5fail/e3ee7aac687880ac72987bb2.png"}, {"type": "image", "uri": "media/ur5fail/2a4ccc470e5f16f76d2421b6.png"}, {"type": "image", "uri": "media/ur5fail/b0cc0ef431e0a58944b41919.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0048/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+1/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 48, "taskvar": "real_put_item_in_drawer+1", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 273, "role": "outcome"}}
{"state": {"task": "put the pink cube in the middle part of the drawer", "subtask": "open middle drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/440e0d84ddd260a6b6e7897b.png"}, {"type": "image", "uri": "media/ur5fail/861f6fc0d855b6aac7b54e44.png"}, {"type": "image", "uri": "media/ur5fail/a9f1366b935a7d647f43261d.png"}, {"type": "image", "uri": "media/ur5fail/e3ee7aac687880ac72987bb2.png"}, {"type": "image", "uri": "media/ur5fail/2a4ccc470e5f16f76d2421b6.png"}, {"type": "image", "uri": "media/ur5fail/b0cc0ef431e0a58944b41919.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0049/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+1/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 49, "taskvar": "real_put_item_in_drawer+1", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the ball, the frog toy, the can and the ball in the top part of the drawer", "subtask": "move top drawer handle to open position", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/5af13b1315f5576216ef7c2e.png"}, {"type": "image", "uri": "media/ur5fail/2957a5df1cc4d101de3277e3.png"}, {"type": "image", "uri": "media/ur5fail/3248863369d59fb2db2d1064.png"}, {"type": "image", "uri": "media/ur5fail/0b2b9b721d004ac700d6219e.png"}, {"type": "image", "uri": "media/ur5fail/47ba64d4311f0a35aefd6ed3.png"}, {"type": "image", "uri": "media/ur5fail/d71fd3da02d52ea12f6f58b9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0050/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 50, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 285, "role": "outcome"}}
{"state": {"task": "put the ball, the frog toy, the can and the ball in the top part of the drawer", "subtask": "move top drawer handle to open position", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/5af13b1315f5576216ef7c2e.png"}, {"type": "image", "uri": "media/ur5fail/2957a5df1cc4d101de3277e3.png"}, {"type": "image", "uri": "media/ur5fail/3248863369d59fb2db2d1064.png"}, {"type": "image", "uri": "media/ur5fail/0d53c757390b99dd6c280646.png"}, {"type": "image", "uri": "media/ur5fail/18d7aeb5fc16d92d1434d42a.png"}, {"type": "image", "uri": "media/ur5fail/1ce21315fa491013b922abae.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0051/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 51, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 285, "role": "outcome"}}
{"state": {"task": "remove the ball, the frog toy, the can and the ball from the top part of the drawer", "subtask": "move top drawer handle to closed position", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/0b2b9b721d004ac700d6219e.png"}, {"type": "image", "uri": "media/ur5fail/47ba64d4311f0a35aefd6ed3.png"}, {"type": "image", "uri": "media/ur5fail/d71fd3da02d52ea12f6f58b9.png"}, {"type": "image", "uri": "media/ur5fail/5af13b1315f5576216ef7c2e.png"}, {"type": "image", "uri": "media/ur5fail/2957a5df1cc4d101de3277e3.png"}, {"type": "image", "uri": "media/ur5fail/3248863369d59fb2db2d1064.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0052/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 52, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 285, "role": "outcome"}}
{"state": {"task": "remove the ball, the frog toy, the can and the ball from the top part of the drawer", "subtask": "move top drawer handle to closed position", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/0b2b9b721d004ac700d6219e.png"}, {"type": "image", "uri": "media/ur5fail/47ba64d4311f0a35aefd6ed3.png"}, {"type": "image", "uri": "media/ur5fail/d71fd3da02d52ea12f6f58b9.png"}, {"type": "image", "uri": "media/ur5fail/ce54c07a64162299e355a37f.png"}, {"type": "image", "uri": "media/ur5fail/bbaa3b960d03a4083d225856.png"}, {"type": "image", "uri": "media/ur5fail/f5c947f6069fde99f129d982.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0053/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 53, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 285, "role": "outcome"}}
{"state": {"task": "put the ball, the frog toy, the can and the ball in the top part of the drawer", "subtask": "move top drawer handle to open position", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/5af13b1315f5576216ef7c2e.png"}, {"type": "image", "uri": "media/ur5fail/2957a5df1cc4d101de3277e3.png"}, {"type": "image", "uri": "media/ur5fail/3248863369d59fb2db2d1064.png"}, {"type": "image", "uri": "media/ur5fail/ce54c07a64162299e355a37f.png"}, {"type": "image", "uri": "media/ur5fail/bbaa3b960d03a4083d225856.png"}, {"type": "image", "uri": "media/ur5fail/f5c947f6069fde99f129d982.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0054/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 54, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 285, "role": "outcome"}}
{"state": {"task": "put the ball, the frog toy, the can and the ball in the top part of the drawer", "subtask": "put the ball in the top part of the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ce54c07a64162299e355a37f.png"}, {"type": "image", "uri": "media/ur5fail/bbaa3b960d03a4083d225856.png"}, {"type": "image", "uri": "media/ur5fail/f5c947f6069fde99f129d982.png"}, {"type": "image", "uri": "media/ur5fail/0b2b9b721d004ac700d6219e.png"}, {"type": "image", "uri": "media/ur5fail/47ba64d4311f0a35aefd6ed3.png"}, {"type": "image", "uri": "media/ur5fail/d71fd3da02d52ea12f6f58b9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0055/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 55, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 288, "role": "outcome"}}
{"state": {"task": "put the frog toy, the ball, the can and the ball in the top part of the drawer", "subtask": "put the frog toy in the top part of the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ce54c07a64162299e355a37f.png"}, {"type": "image", "uri": "media/ur5fail/bbaa3b960d03a4083d225856.png"}, {"type": "image", "uri": "media/ur5fail/f5c947f6069fde99f129d982.png"}, {"type": "image", "uri": "media/ur5fail/64493ea0066bea5625a72999.png"}, {"type": "image", "uri": "media/ur5fail/c1e451812094f56810e45fdb.png"}, {"type": "image", "uri": "media/ur5fail/5d60e4d211d47d13095597c2.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0056/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 56, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 289, "role": "outcome"}}
{"state": {"task": "put the frog toy, the ball, the can and the ball in the top part of the drawer", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d8de0ec0e8c6d654db14a8da.png"}, {"type": "image", "uri": "media/ur5fail/a9d76c79190e7b4d5632d62a.png"}, {"type": "image", "uri": "media/ur5fail/d1bddb1ff41d7035b9125eb4.png"}, {"type": "image", "uri": "media/ur5fail/64493ea0066bea5625a72999.png"}, {"type": "image", "uri": "media/ur5fail/c1e451812094f56810e45fdb.png"}, {"type": "image", "uri": "media/ur5fail/5d60e4d211d47d13095597c2.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0057/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 57, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 283, "role": "outcome"}}
{"state": {"task": "remove the frog toy, the ball, the can and the ball from the top part of the drawer", "subtask": "move grasped object down", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/64493ea0066bea5625a72999.png"}, {"type": "image", "uri": "media/ur5fail/c1e451812094f56810e45fdb.png"}, {"type": "image", "uri": "media/ur5fail/5d60e4d211d47d13095597c2.png"}, {"type": "image", "uri": "media/ur5fail/d8de0ec0e8c6d654db14a8da.png"}, {"type": "image", "uri": "media/ur5fail/a9d76c79190e7b4d5632d62a.png"}, {"type": "image", "uri": "media/ur5fail/d1bddb1ff41d7035b9125eb4.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0058/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 58, "taskvar": "real_put_item_in_drawer+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 283, "role": "outcome"}}
{"state": {"task": "remove the frog toy, the ball, the can and the ball from the top part of the drawer", "subtask": "remove the frog toy from the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/8e6cc293729fdd2f3a21f3f8.png"}, {"type": "image", "uri": "media/ur5fail/00ce687b6c5fddee72f89381.png"}, {"type": "image", "uri": "media/ur5fail/1d16c3eba4d9d7d1a23af5ff.png"}, {"type": "image", "uri": "media/ur5fail/50f0968134a4c492527e101f.png"}, {"type": "image", "uri": "media/ur5fail/10ee984e3c14e96355db4635.png"}, {"type": "image", "uri": "media/ur5fail/92a08cba8b3f23baf961c646.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0059/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 59, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 285, "role": "outcome"}}
{"state": {"task": "put the frog toy, the ball, the can and the ball into the top part of the drawer", "subtask": "move grasped frog toy into the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/50f0968134a4c492527e101f.png"}, {"type": "image", "uri": "media/ur5fail/10ee984e3c14e96355db4635.png"}, {"type": "image", "uri": "media/ur5fail/92a08cba8b3f23baf961c646.png"}, {"type": "image", "uri": "media/ur5fail/8e6cc293729fdd2f3a21f3f8.png"}, {"type": "image", "uri": "media/ur5fail/00ce687b6c5fddee72f89381.png"}, {"type": "image", "uri": "media/ur5fail/1d16c3eba4d9d7d1a23af5ff.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0060/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 60, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 286, "role": "outcome"}}
{"state": {"task": "put the ball, and the pink cube into the top part of the drawer", "subtask": "put the ball into the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/354efcd7a595da712ac03a62.png"}, {"type": "image", "uri": "media/ur5fail/8e31dbd9ed32a6cd94bed111.png"}, {"type": "image", "uri": "media/ur5fail/666637ad904b175114cfa0f7.png"}, {"type": "image", "uri": "media/ur5fail/8e6cc293729fdd2f3a21f3f8.png"}, {"type": "image", "uri": "media/ur5fail/00ce687b6c5fddee72f89381.png"}, {"type": "image", "uri": "media/ur5fail/1d16c3eba4d9d7d1a23af5ff.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0061/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 61, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 279, "role": "outcome"}}
{"state": {"task": "put the ball, and the pink cube into the top part of the drawer", "subtask": "grasp the ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/354efcd7a595da712ac03a62.png"}, {"type": "image", "uri": "media/ur5fail/8e31dbd9ed32a6cd94bed111.png"}, {"type": "image", "uri": "media/ur5fail/666637ad904b175114cfa0f7.png"}, {"type": "image", "uri": "media/ur5fail/50f0968134a4c492527e101f.png"}, {"type": "image", "uri": "media/ur5fail/10ee984e3c14e96355db4635.png"}, {"type": "image", "uri": "media/ur5fail/92a08cba8b3f23baf961c646.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0062/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 62, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 277, "role": "outcome"}}
{"state": {"task": "put the pink cube and the frog toy into the top part of the drawer", "subtask": "put the pink cube in the top part of the drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/43c909e2911aa447e0a6203f.png"}, {"type": "image", "uri": "media/ur5fail/94a49a84c27f44657950d0da.png"}, {"type": "image", "uri": "media/ur5fail/335be5aa8380f4c9bc6f6974.png"}, {"type": "image", "uri": "media/ur5fail/309f2d2c61bfe982e4c393ae.png"}, {"type": "image", "uri": "media/ur5fail/6dfec834cd26a58490d0a0a1.png"}, {"type": "image", "uri": "media/ur5fail/88cf1ace7e0e07b30e7c4105.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0063/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 63, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 284, "role": "outcome"}}
{"state": {"task": "put the pink cube and the frog toy into the middle drawer", "subtask": "open middle drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/43c909e2911aa447e0a6203f.png"}, {"type": "image", "uri": "media/ur5fail/94a49a84c27f44657950d0da.png"}, {"type": "image", "uri": "media/ur5fail/335be5aa8380f4c9bc6f6974.png"}, {"type": "image", "uri": "media/ur5fail/309f2d2c61bfe982e4c393ae.png"}, {"type": "image", "uri": "media/ur5fail/6dfec834cd26a58490d0a0a1.png"}, {"type": "image", "uri": "media/ur5fail/88cf1ace7e0e07b30e7c4105.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0064/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 64, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 273, "role": "outcome"}}
{"state": {"task": "put the pink cube and the frog toy into the middle drawer", "subtask": "open top drawer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/43c909e2911aa447e0a6203f.png"}, {"type": "image", "uri": "media/ur5fail/94a49a84c27f44657950d0da.png"}, {"type": "image", "uri": "media/ur5fail/335be5aa8380f4c9bc6f6974.png"}, {"type": "image", "uri": "media/ur5fail/309f2d2c61bfe982e4c393ae.png"}, {"type": "image", "uri": "media/ur5fail/6dfec834cd26a58490d0a0a1.png"}, {"type": "image", "uri": "media/ur5fail/88cf1ace7e0e07b30e7c4105.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0065/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 65, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 273, "role": "outcome"}}
{"state": {"task": "remove from the top drawer the can, then put into the top drawer the ball, pink cube and frog toy", "subtask": "remove from the top drawer the can", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/646601f30077c7e7b634be8f.png"}, {"type": "image", "uri": "media/ur5fail/d35d19e320c8fd7043e011aa.png"}, {"type": "image", "uri": "media/ur5fail/1e71c2532845c4072bd27c00.png"}, {"type": "image", "uri": "media/ur5fail/5ed9ce08e0e801c87a834cb0.png"}, {"type": "image", "uri": "media/ur5fail/dec6fb5b81e37533be0ce1ab.png"}, {"type": "image", "uri": "media/ur5fail/a8ab708e35c9e6bc4bdf33cf.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0066/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 66, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 287, "role": "outcome"}}
{"state": {"task": "put into the top drawer the frog toy, the ball, and pink cube", "subtask": "put into the top drawer the frog toy", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/b78e1dade212725d9df8e43f.png"}, {"type": "image", "uri": "media/ur5fail/beaf33286a831177023232ff.png"}, {"type": "image", "uri": "media/ur5fail/e89885e8d069f21a55f2c73b.png"}, {"type": "image", "uri": "media/ur5fail/3d7a35a47ea538c9f634c3db.png"}, {"type": "image", "uri": "media/ur5fail/5899ea49179bea304bd1de85.png"}, {"type": "image", "uri": "media/ur5fail/36f88e195bed7223a0e61946.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0067/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 67, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 281, "role": "outcome"}}
{"state": {"task": "put into the top drawer the frog toy, the ball, and pink cube", "subtask": "put into the top drawer the ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/b78e1dade212725d9df8e43f.png"}, {"type": "image", "uri": "media/ur5fail/beaf33286a831177023232ff.png"}, {"type": "image", "uri": "media/ur5fail/e89885e8d069f21a55f2c73b.png"}, {"type": "image", "uri": "media/ur5fail/3d7a35a47ea538c9f634c3db.png"}, {"type": "image", "uri": "media/ur5fail/5899ea49179bea304bd1de85.png"}, {"type": "image", "uri": "media/ur5fail/36f88e195bed7223a0e61946.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0068/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 68, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 280, "role": "outcome"}}
{"state": {"task": "put into the middle drawer the frog toy, the ball, and pink cube", "subtask": "put into the middle drawer the frog toy", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/b78e1dade212725d9df8e43f.png"}, {"type": "image", "uri": "media/ur5fail/beaf33286a831177023232ff.png"}, {"type": "image", "uri": "media/ur5fail/e89885e8d069f21a55f2c73b.png"}, {"type": "image", "uri": "media/ur5fail/3d7a35a47ea538c9f634c3db.png"}, {"type": "image", "uri": "media/ur5fail/5899ea49179bea304bd1de85.png"}, {"type": "image", "uri": "media/ur5fail/36f88e195bed7223a0e61946.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0069/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 69, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 281, "role": "outcome"}}
{"state": {"task": "put into the bottom drawer the frog toy, the ball, and pink cube", "subtask": "put into the bottom drawer the frog toy", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/b78e1dade212725d9df8e43f.png"}, {"type": "image", "uri": "media/ur5fail/beaf33286a831177023232ff.png"}, {"type": "image", "uri": "media/ur5fail/e89885e8d069f21a55f2c73b.png"}, {"type": "image", "uri": "media/ur5fail/3d7a35a47ea538c9f634c3db.png"}, {"type": "image", "uri": "media/ur5fail/5899ea49179bea304bd1de85.png"}, {"type": "image", "uri": "media/ur5fail/36f88e195bed7223a0e61946.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0070/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+3/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 70, "taskvar": "real_put_item_in_drawer+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 281, "role": "outcome"}}
{"state": {"task": "put the ball in the top part of the drawer", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}, {"type": "image", "uri": "media/ur5fail/f8b669e7faa92bc28f1ed854.png"}, {"type": "image", "uri": "media/ur5fail/4ab698304e2d515e243251a1.png"}, {"type": "image", "uri": "media/ur5fail/82a16dd27111676943bb1030.png"}, {"type": "image", "uri": "media/ur5fail/e0c54f3c70e1e8cb996101f4.png"}, {"type": "image", "uri": "media/ur5fail/a491c9f4992bf015e441f42f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0071/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 71, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 271, "role": "outcome"}}
{"state": {"task": "put the ball in the top part of the drawer", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}, {"type": "image", "uri": "media/ur5fail/f8b669e7faa92bc28f1ed854.png"}, {"type": "image", "uri": "media/ur5fail/4ab698304e2d515e243251a1.png"}, {"type": "image", "uri": "media/ur5fail/4794fdd71802ea81ae3880c2.png"}, {"type": "image", "uri": "media/ur5fail/7193e823ccf72ea1de006299.png"}, {"type": "image", "uri": "media/ur5fail/f4e1dc09cd9fceb33116e9d6.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0072/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 72, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 271, "role": "outcome"}}
{"state": {"task": "pick up the fruit between the cylinder and the cube", "subtask": "grasp lemon", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/f7ba6aa0c883551715c77b6d.png"}, {"type": "image", "uri": "media/ur5fail/c7c36a1941252b8c57beda98.png"}, {"type": "image", "uri": "media/ur5fail/d8528fe171162b2de38dd118.png"}, {"type": "image", "uri": "media/ur5fail/280a4237e07f60d8ff8264a3.png"}, {"type": "image", "uri": "media/ur5fail/d200ca8880bd48755a7576af.png"}, {"type": "image", "uri": "media/ur5fail/daa5d3c3360544589d609f3f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0073/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_box+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 73, "taskvar": "real_put_item_spatial_in_box+0", "episode_id": 0, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 271, "role": "outcome"}}
{"state": {"task": "pick up the vegetable", "subtask": "grasp vegetable", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/1745db52a98aced3a79ab81b.png"}, {"type": "image", "uri": "media/ur5fail/00bd895937658778d80e366e.png"}, {"type": "image", "uri": "media/ur5fail/590d61b41d22d86512e24195.png"}, {"type": "image", "uri": "media/ur5fail/ce7fca1bc268dc059331c395.png"}, {"type": "image", "uri": "media/ur5fail/f397295792d19eeb40dfc5a6.png"}, {"type": "image", "uri": "media/ur5fail/af7ca8853b515e5e01404e92.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0074/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817412316", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 74, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817412316, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 265, "role": "outcome"}}
{"state": {"task": "store the tuna can", "subtask": "grasp tuna can", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/97e8fd650d64e2b78f5d67d0.png"}, {"type": "image", "uri": "media/ur5fail/fe32d400c1df09633ec660cf.png"}, {"type": "image", "uri": "media/ur5fail/d5533e45f9f611fdeebc88d9.png"}, {"type": "image", "uri": "media/ur5fail/97e8fd650d64e2b78f5d67d0.png"}, {"type": "image", "uri": "media/ur5fail/fe32d400c1df09633ec660cf.png"}, {"type": "image", "uri": "media/ur5fail/d5533e45f9f611fdeebc88d9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0075/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 75, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 266, "role": "outcome"}}
{"state": {"task": "give me some pizza", "subtask": "grasp pizza", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/30eaecc29d9fef2fdf7cf88c.png"}, {"type": "image", "uri": "media/ur5fail/7e241a0a7b0cb22f9d37b6db.png"}, {"type": "image", "uri": "media/ur5fail/77c170d19d22916f1c6412a4.png"}, {"type": "image", "uri": "media/ur5fail/197204d71a1318eee15ae4b8.png"}, {"type": "image", "uri": "media/ur5fail/6032b5750045c7061f5b945a.png"}, {"type": "image", "uri": "media/ur5fail/af70c51fdf241383312ba777.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0076/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816477261", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 76, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816477261, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 265, "role": "outcome"}}
{"state": {"task": "pick up the strawberry and place it in the blue plate", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/8c8ac4743d6ac066ceb1c36f.png"}, {"type": "image", "uri": "media/ur5fail/4a8f1e0d9f06ef465cbbc7c4.png"}, {"type": "image", "uri": "media/ur5fail/8b819083a3da55bb12b8a835.png"}, {"type": "image", "uri": "media/ur5fail/c654d4badd296298bd66833d.png"}, {"type": "image", "uri": "media/ur5fail/163adc660bb35dd3be6f3c9d.png"}, {"type": "image", "uri": "media/ur5fail/837ed2a5eae03e2092dd20a4.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0077/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816484059", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 77, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816484059, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "grasp the strawberry and place it in the blue plate", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/184cbf96c4115286374071b8.png"}, {"type": "image", "uri": "media/ur5fail/195da17355e6c9aa2675b7b2.png"}, {"type": "image", "uri": "media/ur5fail/573c6584917ffa4cb216f848.png"}, {"type": "image", "uri": "media/ur5fail/f3d5d81fbba5dd52e11044ee.png"}, {"type": "image", "uri": "media/ur5fail/914b224d29a607eb62ec746c.png"}, {"type": "image", "uri": "media/ur5fail/373d34587c1bb680b3f5cc2a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0078/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816492337", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 78, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816492337, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "pick the pear and place it in the container", "subtask": "grasp pear", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/66f558ab24ef777dc5829b87.png"}, {"type": "image", "uri": "media/ur5fail/10165d0fec411d8385c8c656.png"}, {"type": "image", "uri": "media/ur5fail/5c31929d4c59e7ec6b5e6d54.png"}, {"type": "image", "uri": "media/ur5fail/aa7b95d5f56ae99a9ba77c2b.png"}, {"type": "image", "uri": "media/ur5fail/88b874da575ce87531700364.png"}, {"type": "image", "uri": "media/ur5fail/4bfa6b87415402e4fcc85815.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0079/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816518206", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 79, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816518206, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "pick the pear and place it in the container", "subtask": "grasp pear", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/63301fad92e9e8cbbea783c3.png"}, {"type": "image", "uri": "media/ur5fail/aea00e2d23c46b64ac7ea662.png"}, {"type": "image", "uri": "media/ur5fail/b91e8f4e6ff9bf3db8b1a83c.png"}, {"type": "image", "uri": "media/ur5fail/864ad19a1a44505479230936.png"}, {"type": "image", "uri": "media/ur5fail/ca85fb5bbe842b0aba4581e3.png"}, {"type": "image", "uri": "media/ur5fail/10b9fc2bcdf63472804b366d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0080/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816542675", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 80, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816542675, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "pick the ball and place it in the box", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/3a5a74a967478def2803bd97.png"}, {"type": "image", "uri": "media/ur5fail/392946f7dd4736e5d837267e.png"}, {"type": "image", "uri": "media/ur5fail/c458b3278224fb25ebb30866.png"}, {"type": "image", "uri": "media/ur5fail/34705fd86374fa0c0a1833a9.png"}, {"type": "image", "uri": "media/ur5fail/72457308fec03975fa5fd79e.png"}, {"type": "image", "uri": "media/ur5fail/0f05ec355a1dae1e3c1a494b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0081/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816546187", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 81, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816546187, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the box", "subtask": "grasp pear", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ec638e80de8892b12c5ec2dc.png"}, {"type": "image", "uri": "media/ur5fail/f8b9d28d745ad55a0c3777f6.png"}, {"type": "image", "uri": "media/ur5fail/2faa331376241fb59bbfea03.png"}, {"type": "image", "uri": "media/ur5fail/46a025e974f6b82cd32eb267.png"}, {"type": "image", "uri": "media/ur5fail/6ff3d98a1903770f4a176e84.png"}, {"type": "image", "uri": "media/ur5fail/ab48980b4f72561497733a81.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0082/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816550049", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 82, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816550049, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 269, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the box", "subtask": "grasp pear", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/346b0c0621b7ca68f0b77d98.png"}, {"type": "image", "uri": "media/ur5fail/aee2b02ad03ac03a855b3e98.png"}, {"type": "image", "uri": "media/ur5fail/ac9d884305cf8816e342a9d5.png"}, {"type": "image", "uri": "media/ur5fail/406b570430ab1a1bcc8f8118.png"}, {"type": "image", "uri": "media/ur5fail/e34c17ce59a6b9c0ff38a6a3.png"}, {"type": "image", "uri": "media/ur5fail/aeb522b62a2abe3c99534902.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0083/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816553691", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 83, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816553691, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 269, "role": "outcome"}}
{"state": {"task": "grasp a fruit", "subtask": "grasp green pear", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/9815bfc6bf99cde2d7f3ad41.png"}, {"type": "image", "uri": "media/ur5fail/d9165381209ccd4aeb5170f2.png"}, {"type": "image", "uri": "media/ur5fail/764564cb061835c2a481e625.png"}, {"type": "image", "uri": "media/ur5fail/7cf39697f0b5c9022169f780.png"}, {"type": "image", "uri": "media/ur5fail/b24b05f95e35c6a751b2178c.png"}, {"type": "image", "uri": "media/ur5fail/d1006b41e9bdbcaa1bf9a0b0.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0084/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816563503", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 84, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816563503, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 266, "role": "outcome"}}
{"state": {"task": "place in the box the donut", "subtask": "grasp donut", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/ae77aab43a373759ee9c5599.png"}, {"type": "image", "uri": "media/ur5fail/5d916adae8dc9bd8be8024d6.png"}, {"type": "image", "uri": "media/ur5fail/a5aeddc031edb71762cfabfb.png"}, {"type": "image", "uri": "media/ur5fail/38dcd0e6aab390159ad1c277.png"}, {"type": "image", "uri": "media/ur5fail/1a686f8089d46bb574ed56b8.png"}, {"type": "image", "uri": "media/ur5fail/cedf649416d4792f3b266e27.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0085/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816572150", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 85, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816572150, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 269, "role": "outcome"}}
{"state": {"task": "place in the box the chocolate", "subtask": "grasp chocolate", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/98405d9e80f43cbc72e8a360.png"}, {"type": "image", "uri": "media/ur5fail/bc20ebd101bdfec9f84e3a70.png"}, {"type": "image", "uri": "media/ur5fail/40de28d7997e36ee1dea45a0.png"}, {"type": "image", "uri": "media/ur5fail/3b410d76e56d3b9e486feaa8.png"}, {"type": "image", "uri": "media/ur5fail/2ae0a59d6c1423bd8d943230.png"}, {"type": "image", "uri": "media/ur5fail/ec9d08ea87d2fc29698dcbbc.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0086/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816578860", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 86, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816578860, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 267, "role": "outcome"}}
{"state": {"task": "grab the healthiest food", "subtask": "grasp carrot", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/a721c33537bfdecba849aa6d.png"}, {"type": "image", "uri": "media/ur5fail/76053c9b958ad0dc824080b6.png"}, {"type": "image", "uri": "media/ur5fail/d63ec548eb550e75d2b86b74.png"}, {"type": "image", "uri": "media/ur5fail/85883449a03b32f768ada8e4.png"}, {"type": "image", "uri": "media/ur5fail/d5d5c86374d3b9518c8b5745.png"}, {"type": "image", "uri": "media/ur5fail/661ff378c08fc49198d824ea.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0087/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817342016", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 87, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817342016, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 266, "role": "outcome"}}
{"state": {"task": "grab the vegetable", "subtask": "grasp carrot", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/138f5eaf01013f703f20adfe.png"}, {"type": "image", "uri": "media/ur5fail/8e6a7266d896b816989b6c3a.png"}, {"type": "image", "uri": "media/ur5fail/61ae1b3828cc5d60d8ddeb07.png"}, {"type": "image", "uri": "media/ur5fail/5f433983faa030b406ce5dcc.png"}, {"type": "image", "uri": "media/ur5fail/c9ee7884ecb7d1a4fdd851cd.png"}, {"type": "image", "uri": "media/ur5fail/1ae043e196d16142f2067547.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0088/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817351532", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 88, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817351532, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 264, "role": "outcome"}}
{"state": {"task": "grab the bread", "subtask": "grasp bread", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/a5aa0fee63a2954e64f83a81.png"}, {"type": "image", "uri": "media/ur5fail/7aca91b52da15d8e05e9f060.png"}, {"type": "image", "uri": "media/ur5fail/ac2b8647caa7887829ea05fd.png"}, {"type": "image", "uri": "media/ur5fail/3fb27b5269c0db10407969fa.png"}, {"type": "image", "uri": "media/ur5fail/d7e8425c950a735204da7573.png"}, {"type": "image", "uri": "media/ur5fail/ca319868359898b9b0387792.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0089/ur5fail_exec_success", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817368328", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 89, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817368328, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 264, "role": "outcome"}}
{"state": {"task": "put the ball next to the carrot", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/458ccd670843141efd784cbd.png"}, {"type": "image", "uri": "media/ur5fail/ed75defdd6ceb5350129dd89.png"}, {"type": "image", "uri": "media/ur5fail/5f8aaa25f13311be07f8f197.png"}, {"type": "image", "uri": "media/ur5fail/989964cb826e02a9e986ee16.png"}, {"type": "image", "uri": "media/ur5fail/0d00473b2be5733d5ea6c405.png"}, {"type": "image", "uri": "media/ur5fail/77b19a7e954a9904760cd52a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0090/ur5fail_exec_success", "group_id": "ur5fail/real_sport+0/173817680775", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 90, "taskvar": "real_sport+0", "episode_id": 173817680775, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 268, "role": "outcome"}}
{"state": {"task": "put the ball next to the carrot", "subtask": "grasp ball", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/458ccd670843141efd784cbd.png"}, {"type": "image", "uri": "media/ur5fail/ed75defdd6ceb5350129dd89.png"}, {"type": "image", "uri": "media/ur5fail/5f8aaa25f13311be07f8f197.png"}, {"type": "image", "uri": "media/ur5fail/8a5146e75c28feb7b76a1f13.png"}, {"type": "image", "uri": "media/ur5fail/125a62d2b87f4c6f687225fe.png"}, {"type": "image", "uri": "media/ur5fail/5bde157afd9bcc21cd619d7e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0091/ur5fail_exec_success", "group_id": "ur5fail/real_sport+0/173817680775", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 91, "taskvar": "real_sport+0", "episode_id": 173817680775, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 268, "role": "outcome"}}
{"state": {"task": "pick up a fruit", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/c31047ac4721aed2f8638571.png"}, {"type": "image", "uri": "media/ur5fail/dd4bbd90e99f682b1ead0806.png"}, {"type": "image", "uri": "media/ur5fail/1a4fb7ba01b1e69ddd18ece6.png"}, {"type": "image", "uri": "media/ur5fail/1990a4d607f6a1d2ff243ffc.png"}, {"type": "image", "uri": "media/ur5fail/f250751a7629f976a0704f2f.png"}, {"type": "image", "uri": "media/ur5fail/761afeff6baadd3fc5a8083b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0092/ur5fail_exec_success", "group_id": "ur5fail/real_sport+0/173817657492", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 92, "taskvar": "real_sport+0", "episode_id": 173817657492, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 265, "role": "outcome"}}
{"state": {"task": "stack a black cup onto a red cup", "subtask": "grasp black cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/81633cb1b3536fc48d1501ee.png"}, {"type": "image", "uri": "media/ur5fail/06c3b29a69c09d25446e0229.png"}, {"type": "image", "uri": "media/ur5fail/c295e8c91be57af41942950c.png"}, {"type": "image", "uri": "media/ur5fail/028c93cd702b9ed2ff432a27.png"}, {"type": "image", "uri": "media/ur5fail/2263de9fc61e49c4ead1b9e6.png"}, {"type": "image", "uri": "media/ur5fail/baa0cf41a9fc70c92f56a335.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0093/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 93, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "stack a red cup onto a black cup", "subtask": "grasp red cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/5d9d68cfbc7ffb04d87e6670.png"}, {"type": "image", "uri": "media/ur5fail/f5f0e25d26ba93d0cd5e395d.png"}, {"type": "image", "uri": "media/ur5fail/2192139fb1a363bbd9a0c64c.png"}, {"type": "image", "uri": "media/ur5fail/cc3704cda2750c39c00087f7.png"}, {"type": "image", "uri": "media/ur5fail/fa4cd4d9113344cc3f0d2836.png"}, {"type": "image", "uri": "media/ur5fail/5a30a46bdadef7ad784fa5da.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0094/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 94, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "stack onto the cyan cup the cup that is the farthest away from the cyan cup", "subtask": "grasp the cup that is the farthest away from the cyan cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/5d9d68cfbc7ffb04d87e6670.png"}, {"type": "image", "uri": "media/ur5fail/f5f0e25d26ba93d0cd5e395d.png"}, {"type": "image", "uri": "media/ur5fail/2192139fb1a363bbd9a0c64c.png"}, {"type": "image", "uri": "media/ur5fail/cc3704cda2750c39c00087f7.png"}, {"type": "image", "uri": "media/ur5fail/fa4cd4d9113344cc3f0d2836.png"}, {"type": "image", "uri": "media/ur5fail/5a30a46bdadef7ad784fa5da.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0095/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 95, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 289, "role": "outcome"}}
{"state": {"task": "put a red cup onto a yellow cup", "subtask": "grasp red cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/70ffc01d9165fa32c95d1877.png"}, {"type": "image", "uri": "media/ur5fail/ed42faef6cc3ade60794c25e.png"}, {"type": "image", "uri": "media/ur5fail/97e816a68d830218da0db357.png"}, {"type": "image", "uri": "media/ur5fail/d5bbd2cbfed3db07f9b75bff.png"}, {"type": "image", "uri": "media/ur5fail/9ca84f61b5d9348fbcc7df66.png"}, {"type": "image", "uri": "media/ur5fail/adf441509ca035aa6bb8e9d8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0096/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 96, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "stack black cup on orange cup", "subtask": "move grasped object to orange cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/72881107a1c7da71968b49f5.png"}, {"type": "image", "uri": "media/ur5fail/1e9e79caca4c791d3cf14be8.png"}, {"type": "image", "uri": "media/ur5fail/11a9728451260881ea8b1238.png"}, {"type": "image", "uri": "media/ur5fail/96d7519ee92bb985ad893253.png"}, {"type": "image", "uri": "media/ur5fail/26c689eee8ffdae20d8c469f.png"}, {"type": "image", "uri": "media/ur5fail/73a5b5a3ffd3ac2cf7635128.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0097/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+5/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 97, "taskvar": "real_stack_cup+5", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 271, "role": "outcome"}}
{"state": {"task": "unstack the cups", "subtask": "move grasped object to table", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/96d7519ee92bb985ad893253.png"}, {"type": "image", "uri": "media/ur5fail/26c689eee8ffdae20d8c469f.png"}, {"type": "image", "uri": "media/ur5fail/73a5b5a3ffd3ac2cf7635128.png"}, {"type": "image", "uri": "media/ur5fail/72881107a1c7da71968b49f5.png"}, {"type": "image", "uri": "media/ur5fail/1e9e79caca4c791d3cf14be8.png"}, {"type": "image", "uri": "media/ur5fail/11a9728451260881ea8b1238.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0098/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+5/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 98, "taskvar": "real_stack_cup+5", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 268, "role": "outcome"}}
{"state": {"task": "stack black cup on orange cup", "subtask": "grasp black cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/c24d8e97cccbe22ab223bb25.png"}, {"type": "image", "uri": "media/ur5fail/9e75b87b143305195f656823.png"}, {"type": "image", "uri": "media/ur5fail/5c45fb2ccf71083ee6f06165.png"}, {"type": "image", "uri": "media/ur5fail/e93fc4e84288204d8373f8c2.png"}, {"type": "image", "uri": "media/ur5fail/06f0eed14e3bff6a16ff9040.png"}, {"type": "image", "uri": "media/ur5fail/80b6c36a7f1456cf4f001ceb.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0099/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+5/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 99, "taskvar": "real_stack_cup+5", "episode_id": 0, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 268, "role": "outcome"}}
{"state": {"task": "stack black cup on orange cup", "subtask": "grasp black cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/c81457951a497ce999dac720.png"}, {"type": "image", "uri": "media/ur5fail/814371a5fc50db611d012e97.png"}, {"type": "image", "uri": "media/ur5fail/03c8628711a01c6a4e0545db.png"}, {"type": "image", "uri": "media/ur5fail/1515e5a41fa363217a925fe5.png"}, {"type": "image", "uri": "media/ur5fail/9730c66a1e5e1bca19b9ab2e.png"}, {"type": "image", "uri": "media/ur5fail/f07297b79868a48d81f551f1.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0100/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+5/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 100, "taskvar": "real_stack_cup+5", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 268, "role": "outcome"}}
{"state": {"task": "stack orange cup on top of the cup that is the closest to the red apple", "subtask": "grasp orange cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/6af2e641d831d29db4c016c2.png"}, {"type": "image", "uri": "media/ur5fail/1f318a98974cdc6c32061564.png"}, {"type": "image", "uri": "media/ur5fail/3e727af345a412465317c183.png"}, {"type": "image", "uri": "media/ur5fail/b7170b7f692bcb47351b9a27.png"}, {"type": "image", "uri": "media/ur5fail/4d5cfafbcba6fa274d8af150.png"}, {"type": "image", "uri": "media/ur5fail/7913f10646cb9ed6f70af377.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0101/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup_spatial+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 101, "taskvar": "real_stack_cup_spatial+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 278, "role": "outcome"}}
{"state": {"task": "stack onto the yellow cup the cup that is the farthest from the red apple", "subtask": "grasp the cup that is the farthest from the red apple", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/6af2e641d831d29db4c016c2.png"}, {"type": "image", "uri": "media/ur5fail/1f318a98974cdc6c32061564.png"}, {"type": "image", "uri": "media/ur5fail/3e727af345a412465317c183.png"}, {"type": "image", "uri": "media/ur5fail/b7170b7f692bcb47351b9a27.png"}, {"type": "image", "uri": "media/ur5fail/4d5cfafbcba6fa274d8af150.png"}, {"type": "image", "uri": "media/ur5fail/7913f10646cb9ed6f70af377.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0102/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup_spatial+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 102, "taskvar": "real_stack_cup_spatial+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 287, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box, then put the corn in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/8776f1f468a3e413d689ccb8.png"}, {"type": "image", "uri": "media/ur5fail/c6cef417076f1284bc361dec.png"}, {"type": "image", "uri": "media/ur5fail/efa05d0108a0333df8ed5135.png"}, {"type": "image", "uri": "media/ur5fail/1fdde352d142659373f2c9e0.png"}, {"type": "image", "uri": "media/ur5fail/bf63521c281a4e6a283c6e2f.png"}, {"type": "image", "uri": "media/ur5fail/27582eaab1e8c55a25632ef7.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0103/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 103, "taskvar": "real_put_food_in_box+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 279, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box, then put the corn in the box", "subtask": "grasp corn", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/4c1ee0712f052cac143685ad.png"}, {"type": "image", "uri": "media/ur5fail/6787346e8ad2e8d90da97783.png"}, {"type": "image", "uri": "media/ur5fail/1b49c2ae66eb59c71ac5730b.png"}, {"type": "image", "uri": "media/ur5fail/5a402851e9708bfd1091cb75.png"}, {"type": "image", "uri": "media/ur5fail/6595ae26f243c85ebf2a2f37.png"}, {"type": "image", "uri": "media/ur5fail/9be3bc5cf8a56bca7f764edf.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0104/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 104, "taskvar": "real_put_food_in_box+0", "episode_id": 2, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 276, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box, then put the corn in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/5a402851e9708bfd1091cb75.png"}, {"type": "image", "uri": "media/ur5fail/6595ae26f243c85ebf2a2f37.png"}, {"type": "image", "uri": "media/ur5fail/9be3bc5cf8a56bca7f764edf.png"}, {"type": "image", "uri": "media/ur5fail/c1f966e1d9cd5f45714d9e17.png"}, {"type": "image", "uri": "media/ur5fail/85d7e6cecc8d7f10bd3f26ed.png"}, {"type": "image", "uri": "media/ur5fail/eb48dc4c059f44110cc997fa.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0105/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 105, "taskvar": "real_put_food_in_box+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 279, "role": "outcome"}}
{"state": {"task": "put the strawberry in the box then put the banana in the box", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/21133fdf067010dcfc8dee38.png"}, {"type": "image", "uri": "media/ur5fail/e730ed4442068de5bd5ffaff.png"}, {"type": "image", "uri": "media/ur5fail/7737934e674f58110695df81.png"}, {"type": "image", "uri": "media/ur5fail/9aac2e913179339d614a9d0a.png"}, {"type": "image", "uri": "media/ur5fail/c3bbeadcbc5a96e9675b8153.png"}, {"type": "image", "uri": "media/ur5fail/c36bd1b525809d0ff6f81ce5.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0106/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 106, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 274, "role": "outcome"}}
{"state": {"task": "put the strawberry in the box then put the banana in the box", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/21133fdf067010dcfc8dee38.png"}, {"type": "image", "uri": "media/ur5fail/e730ed4442068de5bd5ffaff.png"}, {"type": "image", "uri": "media/ur5fail/7737934e674f58110695df81.png"}, {"type": "image", "uri": "media/ur5fail/2170201391335c830c06c587.png"}, {"type": "image", "uri": "media/ur5fail/967d242912eb1de47507765c.png"}, {"type": "image", "uri": "media/ur5fail/997a2e84e5da9aa3dd81a748.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0107/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 107, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 274, "role": "outcome"}}
{"state": {"task": "put the strawberry in the box then put the banana in the box", "subtask": "grasp strawberry", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/096e7fabaf984d3c8ef15c9f.png"}, {"type": "image", "uri": "media/ur5fail/42d41d0c7fa7fafa7177ff31.png"}, {"type": "image", "uri": "media/ur5fail/eb9f966c8119a8716c55dc47.png"}, {"type": "image", "uri": "media/ur5fail/d013101b9146b81f4a9a1667.png"}, {"type": "image", "uri": "media/ur5fail/bc35e990b01c83c43a39a5fd.png"}, {"type": "image", "uri": "media/ur5fail/cc4ad03ff36233b3254872ec.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0108/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 108, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 274, "role": "outcome"}}
{"state": {"task": "put the banana in the box then put the lemon in the box", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/6cef5ecf60ac0402167aed52.png"}, {"type": "image", "uri": "media/ur5fail/733dd6cbf433ec527e253797.png"}, {"type": "image", "uri": "media/ur5fail/8b54a697c704e8bbaa01741b.png"}, {"type": "image", "uri": "media/ur5fail/353828c4e2b2f166f915f39f.png"}, {"type": "image", "uri": "media/ur5fail/2ae7ad48899197e77eab74cf.png"}, {"type": "image", "uri": "media/ur5fail/d4018ce4dfa2fe94ebc7bb40.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0109/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/4", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 109, "taskvar": "real_put_food_in_box+0", "episode_id": 4, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 274, "role": "outcome"}}
{"state": {"task": "put the banana in the box then put the lemon in the box", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/881712c4b262cf35cc807349.png"}, {"type": "image", "uri": "media/ur5fail/f9df2eebabb1a1e1d8d228c4.png"}, {"type": "image", "uri": "media/ur5fail/ca0c296c82d46a88e62869e6.png"}, {"type": "image", "uri": "media/ur5fail/a9f5a369ee3079814d88b55f.png"}, {"type": "image", "uri": "media/ur5fail/50d9e7bab2bd23f461c79e01.png"}, {"type": "image", "uri": "media/ur5fail/e15ee26f6d30d14e946f23de.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0110/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/5", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 110, "taskvar": "real_put_food_in_box+0", "episode_id": 5, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 274, "role": "outcome"}}
{"state": {"task": "pick the pink cup and put it onto the cyan cup", "subtask": "grasp pink cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d8cd563714ee4dc92707a752.png"}, {"type": "image", "uri": "media/ur5fail/ed168a91fc834eb224b84f1d.png"}, {"type": "image", "uri": "media/ur5fail/46bf29ec1935e2dbe6dc3d6b.png"}, {"type": "image", "uri": "media/ur5fail/59b0e58f747668c010cac56a.png"}, {"type": "image", "uri": "media/ur5fail/42d10d70b33c8ff069677e01.png"}, {"type": "image", "uri": "media/ur5fail/59d65e104b9cb5f48c6482b5.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0111/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cups+2/10", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 111, "taskvar": "real_stack_cups+2", "episode_id": 10, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 273, "role": "outcome"}}
{"state": {"task": "put the tuna can and the corn in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/72dc24791f92e14e3bc9a830.png"}, {"type": "image", "uri": "media/ur5fail/1470dfdb76c2e953cc138238.png"}, {"type": "image", "uri": "media/ur5fail/114c40a2d57b38055a23d2e6.png"}, {"type": "image", "uri": "media/ur5fail/be29a32d07cdf769af3bf893.png"}, {"type": "image", "uri": "media/ur5fail/3040259c04951b6e85fee5c4.png"}, {"type": "image", "uri": "media/ur5fail/088dc48b423b0f2e9018788a.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0112/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 112, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 274, "role": "outcome"}}
{"state": {"task": "put the green fruit in the box", "subtask": "grasp green pear", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/e5057a0d51fbdbd3ed4948c7.png"}, {"type": "image", "uri": "media/ur5fail/0ab870b13ff9c60649b6602f.png"}, {"type": "image", "uri": "media/ur5fail/ad270b30fb9464d4ab3037eb.png"}, {"type": "image", "uri": "media/ur5fail/0862f4ad24023fb2f675e5cd.png"}, {"type": "image", "uri": "media/ur5fail/bd36cfe877ace0d589546739.png"}, {"type": "image", "uri": "media/ur5fail/d7a58b8d7ab988568b368555.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0113/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 113, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 269, "role": "outcome"}}
{"state": {"task": "put another fruit in the box with the strawberry", "subtask": "put red apple in box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/967006959e9808bb4e9d3b75.png"}, {"type": "image", "uri": "media/ur5fail/56e92cfba96e2b579df84f93.png"}, {"type": "image", "uri": "media/ur5fail/3f49887a41c3658c9e54801b.png"}, {"type": "image", "uri": "media/ur5fail/bb18c321386820619bd331fc.png"}, {"type": "image", "uri": "media/ur5fail/a7b018f138f5e8830afa3b39.png"}, {"type": "image", "uri": "media/ur5fail/8e8031a8bb75965679423f8d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0114/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/12", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 114, "taskvar": "real_put_food_in_box+0", "episode_id": 12, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the fruits in the box", "subtask": "move grasped object to box", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/8b59dbbea44ae020f107c2e5.png"}, {"type": "image", "uri": "media/ur5fail/87b888552c05c95c2dff927c.png"}, {"type": "image", "uri": "media/ur5fail/fb2ae8573c8b66ad4fc155b1.png"}, {"type": "image", "uri": "media/ur5fail/3181712a0228c1d988a523ad.png"}, {"type": "image", "uri": "media/ur5fail/621c2d1ac78d338765f10b4c.png"}, {"type": "image", "uri": "media/ur5fail/3f4d03eeefdd5007e69e0321.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0115/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/12", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 115, "taskvar": "real_put_food_in_box+0", "episode_id": 12, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "translation_target", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the fruits in the box", "subtask": "move grasped object up", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/8b59dbbea44ae020f107c2e5.png"}, {"type": "image", "uri": "media/ur5fail/87b888552c05c95c2dff927c.png"}, {"type": "image", "uri": "media/ur5fail/fb2ae8573c8b66ad4fc155b1.png"}, {"type": "image", "uri": "media/ur5fail/3181712a0228c1d988a523ad.png"}, {"type": "image", "uri": "media/ur5fail/621c2d1ac78d338765f10b4c.png"}, {"type": "image", "uri": "media/ur5fail/3f4d03eeefdd5007e69e0321.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0116/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/12", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 116, "taskvar": "real_put_food_in_box+0", "episode_id": 12, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 269, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box with the strawberry", "subtask": "grasp tuna can", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/c912b9da87b713654190e29f.png"}, {"type": "image", "uri": "media/ur5fail/22fb2565f53d341ac3666fcd.png"}, {"type": "image", "uri": "media/ur5fail/dea2c3cd84185528dc5e7b98.png"}, {"type": "image", "uri": "media/ur5fail/049c1bab68a1672f379e1bd2.png"}, {"type": "image", "uri": "media/ur5fail/39df22c7d127d77df91b2bdf.png"}, {"type": "image", "uri": "media/ur5fail/c414b8d61aa51367efe3f4f1.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0117/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/12", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 117, "taskvar": "real_put_food_in_box+0", "episode_id": 12, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the tuna can in the box with the strawberry", "subtask": "grasp tuna can", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/c912b9da87b713654190e29f.png"}, {"type": "image", "uri": "media/ur5fail/22fb2565f53d341ac3666fcd.png"}, {"type": "image", "uri": "media/ur5fail/dea2c3cd84185528dc5e7b98.png"}, {"type": "image", "uri": "media/ur5fail/bb18c321386820619bd331fc.png"}, {"type": "image", "uri": "media/ur5fail/a7b018f138f5e8830afa3b39.png"}, {"type": "image", "uri": "media/ur5fail/8e8031a8bb75965679423f8d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0118/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/12", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 118, "taskvar": "real_put_food_in_box+0", "episode_id": 12, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put corn in the box with the tuna can and peach", "subtask": "grasp corn", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/a1b518f5308fb61dc4aaff08.png"}, {"type": "image", "uri": "media/ur5fail/b4662d4bd0b8146a826cab4f.png"}, {"type": "image", "uri": "media/ur5fail/085d780be994d28aa55d63e3.png"}, {"type": "image", "uri": "media/ur5fail/a001f41b7329f074400e1cad.png"}, {"type": "image", "uri": "media/ur5fail/ae9efc31b09e97e57f58c267.png"}, {"type": "image", "uri": "media/ur5fail/c03a46016863c5d847c097f3.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0119/ur5fail_exec_success", "group_id": "ur5fail/real_put_food_in_box+0/13", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 119, "taskvar": "real_put_food_in_box+0", "episode_id": 13, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "my house is broken, please fix it", "subtask": "grasp hammer", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/4f6496608f3b9e7093297875.png"}, {"type": "image", "uri": "media/ur5fail/2fa7ac7648dcb09b0aaadd47.png"}, {"type": "image", "uri": "media/ur5fail/52a3a7c67b8d2c3a81cdd326.png"}, {"type": "image", "uri": "media/ur5fail/4f6496608f3b9e7093297875.png"}, {"type": "image", "uri": "media/ur5fail/2fa7ac7648dcb09b0aaadd47.png"}, {"type": "image", "uri": "media/ur5fail/52a3a7c67b8d2c3a81cdd326.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0120/ur5fail_exec_success", "group_id": "ur5fail/real_pick_tool_gondola+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 120, "taskvar": "real_pick_tool_gondola+0", "episode_id": 0, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 269, "role": "outcome"}}
{"state": {"task": "my house is broken, please fix it", "subtask": "grasp screwdriver", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/21abe77275adebd79bb8b91e.png"}, {"type": "image", "uri": "media/ur5fail/544a6a8490097e7db186ace7.png"}, {"type": "image", "uri": "media/ur5fail/fd6c35e1e15803f1939db35d.png"}, {"type": "image", "uri": "media/ur5fail/21abe77275adebd79bb8b91e.png"}, {"type": "image", "uri": "media/ur5fail/544a6a8490097e7db186ace7.png"}, {"type": "image", "uri": "media/ur5fail/fd6c35e1e15803f1939db35d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0121/ur5fail_exec_success", "group_id": "ur5fail/real_pick_tool_gondola+0/1", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 121, "taskvar": "real_pick_tool_gondola+0", "episode_id": 1, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "clean the table", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/47ff4890929bdb729765135b.png"}, {"type": "image", "uri": "media/ur5fail/3a92b01cea46bc99b2babf9b.png"}, {"type": "image", "uri": "media/ur5fail/3517a8eef2e7a32724dcc07b.png"}, {"type": "image", "uri": "media/ur5fail/47a471f63506a3a5809b5283.png"}, {"type": "image", "uri": "media/ur5fail/a508ad2c957188f4a5d5c1a0.png"}, {"type": "image", "uri": "media/ur5fail/3aa35972390d6f2e8c72a0bf.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0122/ur5fail_exec_success", "group_id": "ur5fail/real_put_carrot_in_container+0/1", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 122, "taskvar": "real_put_carrot_in_container+0", "episode_id": 1, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 264, "role": "outcome"}}
{"state": {"task": "clean the table by putting things in the small box", "subtask": "move grasped object to small container", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/eeaaf69cdcd33cba3e056e62.png"}, {"type": "image", "uri": "media/ur5fail/832d3784e19beef74acd473e.png"}, {"type": "image", "uri": "media/ur5fail/8f1ae6cf12b78a3e480419e0.png"}, {"type": "image", "uri": "media/ur5fail/6e5913725fb1f44939a26f89.png"}, {"type": "image", "uri": "media/ur5fail/be3791c13aa3a2e4f93ba6ed.png"}, {"type": "image", "uri": "media/ur5fail/8591a1732e75b4b83ecb271b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0123/ur5fail_exec_success", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 123, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "slip", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 275, "role": "outcome"}}
{"state": {"task": "put everything in the big box", "subtask": "grasp fruit", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/2368503e919d3001e52b1354.png"}, {"type": "image", "uri": "media/ur5fail/8e53800ca67452243e8188e5.png"}, {"type": "image", "uri": "media/ur5fail/8869d9904265d82d251997d0.png"}, {"type": "image", "uri": "media/ur5fail/4ffeb2fbc92eb6de336550b7.png"}, {"type": "image", "uri": "media/ur5fail/037da0bef756d0df08b55ec5.png"}, {"type": "image", "uri": "media/ur5fail/07fe3b58f2ab85993bfe4a11.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0124/ur5fail_exec_success", "group_id": "ur5fail/real_put_carrot_in_container+0/1", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 124, "taskvar": "real_put_carrot_in_container+0", "episode_id": 1, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 267, "role": "outcome"}}
{"state": {"task": "put everything in the big box", "subtask": "grasp fruit", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}, {"type": "image", "uri": "media/ur5fail/d8a71acbfa3bafd2b32d8c8c.png"}, {"type": "image", "uri": "media/ur5fail/f1207aac490261d7754d08d4.png"}, {"type": "image", "uri": "media/ur5fail/f7e289349d199be148c1fdbb.png"}, {"type": "image", "uri": "media/ur5fail/fb760c1210fa63750a761b9c.png"}, {"type": "image", "uri": "media/ur5fail/9dc053cf22541d79f7dd2648.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0125/ur5fail_exec_success", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 125, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 267, "role": "outcome"}}
{"state": {"task": "put everything in the big box", "subtask": "grasp fruit", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}, {"type": "image", "uri": "media/ur5fail/d8a71acbfa3bafd2b32d8c8c.png"}, {"type": "image", "uri": "media/ur5fail/f1207aac490261d7754d08d4.png"}, {"type": "image", "uri": "media/ur5fail/2cf8d16550fbaf6a43ce23c8.png"}, {"type": "image", "uri": "media/ur5fail/4da1fe1d16e7c3c9141052e8.png"}, {"type": "image", "uri": "media/ur5fail/a7c33c60c69a0f893dcf0750.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0126/ur5fail_exec_success", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 126, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 267, "role": "outcome"}}
{"state": {"task": "clean the things on the table out of the table", "subtask": "grasp white plate", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/5af5193de747ee3e8789b355.png"}, {"type": "image", "uri": "media/ur5fail/2a63444c2fd1219ccb39f44c.png"}, {"type": "image", "uri": "media/ur5fail/6c31269155ea4856245e6ef0.png"}, {"type": "image", "uri": "media/ur5fail/993a535ff8cee0329ba6ce82.png"}, {"type": "image", "uri": "media/ur5fail/e59855253a9adf9a185c7b08.png"}, {"type": "image", "uri": "media/ur5fail/ce3f174a3324600176291949.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0127/ur5fail_exec_success", "group_id": "ur5fail/real_put_carrot_in_plate+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 127, "taskvar": "real_put_carrot_in_plate+0", "episode_id": 0, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "stack the pink cup onto the black cup then the black cup onto the cyan cup, finally stack the cyan cup onto the yellow cup", "subtask": "grasp pink cup", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/e4e68536c5d77ddb2f0cb3fd.png"}, {"type": "image", "uri": "media/ur5fail/894e112a24b9eb34c100bd17.png"}, {"type": "image", "uri": "media/ur5fail/dcb4ab8b1d35b4a3c685eb00.png"}, {"type": "image", "uri": "media/ur5fail/908bba46dd5f97b1e6483706.png"}, {"type": "image", "uri": "media/ur5fail/36cbd115b11d8caff7f7d75e.png"}, {"type": "image", "uri": "media/ur5fail/6a4b540082ad4fa99447a590.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0128/ur5fail_exec_success", "group_id": "ur5fail/real_stack_cup+3/5", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 128, "taskvar": "real_stack_cup+3", "episode_id": 5, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 288, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/db912aa59345d929bbd360c8.png"}, {"type": "image", "uri": "media/ur5fail/58ace0d98d14e15c58f29c00.png"}, {"type": "image", "uri": "media/ur5fail/876666ea98e6c485dda1fc0b.png"}, {"type": "image", "uri": "media/ur5fail/deff1af6cd9bb45e4629350c.png"}, {"type": "image", "uri": "media/ur5fail/f5cb388c8b8c209b171ede93.png"}, {"type": "image", "uri": "media/ur5fail/1ed7f4a50c2c350eb7e8cf94.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0129/ur5fail_exec_success", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 129, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 2, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d32b55e6429bafd98990271a.png"}, {"type": "image", "uri": "media/ur5fail/049bd6cfdb13be69a1b2dbba.png"}, {"type": "image", "uri": "media/ur5fail/4f7869f3274b449e0c26f8c8.png"}, {"type": "image", "uri": "media/ur5fail/5de64c7e5f3d62a2e777fd8e.png"}, {"type": "image", "uri": "media/ur5fail/1dd95f753835f5c5aad10c57.png"}, {"type": "image", "uri": "media/ur5fail/98dbed86a1523ae2d6ac5a10.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0130/ur5fail_exec_success", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/3", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 130, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 3, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the pink plate on top of the navy plate, and move the healthiest food to the pink plate", "subtask": "grasp pink plate", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/aa26ac7fe092ff0e6d5b9978.png"}, {"type": "image", "uri": "media/ur5fail/9de7667295a8a01c76bf0a34.png"}, {"type": "image", "uri": "media/ur5fail/0795bd2a268cf77a48ccf76c.png"}, {"type": "image", "uri": "media/ur5fail/79148c63aeaa0b1326635ed5.png"}, {"type": "image", "uri": "media/ur5fail/81feb4854ee70b7bcb53a456.png"}, {"type": "image", "uri": "media/ur5fail/8221e636992774bb6433aa89.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0131/ur5fail_exec_success", "group_id": "ur5fail/real_healthiest_banana_in_plate+1/7", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 131, "taskvar": "real_healthiest_banana_in_plate+1", "episode_id": 7, "failure_category": "imprecise_grasp_or_push", "failure_mode_raw": "translation_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 283, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp banana", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/d16305d9827d7935065b5e65.png"}, {"type": "image", "uri": "media/ur5fail/cfbd22e5c0fe2cd4db3bcf75.png"}, {"type": "image", "uri": "media/ur5fail/c7fec6ac21d39dfdd22e153e.png"}, {"type": "image", "uri": "media/ur5fail/866394b9585dbc9cf2064bc1.png"}, {"type": "image", "uri": "media/ur5fail/538d96d4b960334a840b5dfc.png"}, {"type": "image", "uri": "media/ur5fail/4049c07fbd2eeab3d7afae70.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0132/ur5fail_exec_success", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/8", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 132, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 8, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 270, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp healthiest food", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/2cc4c67d22468ce90cd65273.png"}, {"type": "image", "uri": "media/ur5fail/02a24eba277a34207592757a.png"}, {"type": "image", "uri": "media/ur5fail/3feafbe07ce3834b8c26418a.png"}, {"type": "image", "uri": "media/ur5fail/caeb7759f2f0a0aa7a80a73c.png"}, {"type": "image", "uri": "media/ur5fail/dabb349694fe01913ebf8036.png"}, {"type": "image", "uri": "media/ur5fail/39a62b0fcf7b854244d0612e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0133/ur5fail_exec_success", "group_id": "ur5fail/real_healthiest_in_plate+0/0", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 133, "taskvar": "real_healthiest_in_plate+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong_object", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp healthiest food", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/935a5d8bf30047930c7b31f9.png"}, {"type": "image", "uri": "media/ur5fail/fd33c437583ef1723d2aa593.png"}, {"type": "image", "uri": "media/ur5fail/c1075da5b518204d2ee81ae9.png"}, {"type": "image", "uri": "media/ur5fail/ca937ceff168553bfb6e64ec.png"}, {"type": "image", "uri": "media/ur5fail/1cf28175b0d5a51338306619.png"}, {"type": "image", "uri": "media/ur5fail/518849293fb1f54468bf65a1.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0134/ur5fail_exec_success", "group_id": "ur5fail/real_healthiest_in_plate+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 134, "taskvar": "real_healthiest_in_plate+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp healthiest food", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/a57594f270ccd1eaac281bf9.png"}, {"type": "image", "uri": "media/ur5fail/cb5482b5e8cc60822d283e59.png"}, {"type": "image", "uri": "media/ur5fail/14bab0f5d41a55d50c134cab.png"}, {"type": "image", "uri": "media/ur5fail/b9d2f6180d4e3574f36808c0.png"}, {"type": "image", "uri": "media/ur5fail/8be278f95135c95260f8ea83.png"}, {"type": "image", "uri": "media/ur5fail/d564f17b5f20bfde28681572.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0135/ur5fail_exec_success", "group_id": "ur5fail/real_healthiest_in_plate+0/3", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 135, "taskvar": "real_healthiest_in_plate+0", "episode_id": 3, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the healthiest food in the pink plate", "subtask": "grasp healthiest food", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/6abd8f8fa1b9d6d5298c7f4b.png"}, {"type": "image", "uri": "media/ur5fail/c60342d932eaa5b210f76c58.png"}, {"type": "image", "uri": "media/ur5fail/af8f3bacdae56920f76556d7.png"}, {"type": "image", "uri": "media/ur5fail/54c26ae7eaf665f03d6a8570.png"}, {"type": "image", "uri": "media/ur5fail/e9dffebea2d744c04e15af69.png"}, {"type": "image", "uri": "media/ur5fail/765358a293c7e8599eacaf0c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0136/ur5fail_exec_success", "group_id": "ur5fail/real_healthiest_in_plate+0/5", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 136, "taskvar": "real_healthiest_in_plate+0", "episode_id": 5, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 272, "role": "outcome"}}
{"state": {"task": "put the fruit in the plate on the napkin", "subtask": "grasp fruit", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}, {"type": "image", "uri": "media/ur5fail/76b9427d2847a0cf5497ac2b.png"}, {"type": "image", "uri": "media/ur5fail/74be04cbf1890036dfef742f.png"}, {"type": "image", "uri": "media/ur5fail/bda84355137835a947492757.png"}, {"type": "image", "uri": "media/ur5fail/016c87a5724c42f6182919c3.png"}, {"type": "image", "uri": "media/ur5fail/467b9435b6a84a2507d9a6cc.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0137/ur5fail_exec_success", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 137, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 271, "role": "outcome"}}
{"state": {"task": "put the fruit in the plate on the napkin", "subtask": "grasp fruit", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": true, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/14cf03f49f8ca33fb67686a8.png"}, {"type": "image", "uri": "media/ur5fail/4c989aa9b6156dd23cc0b356.png"}, {"type": "image", "uri": "media/ur5fail/31452e3c0d54c18f6f1c55d3.png"}, {"type": "image", "uri": "media/ur5fail/8be9c8c62f1ea40e3201acb3.png"}, {"type": "image", "uri": "media/ur5fail/3b8e66ec7c0ae69de1796da2.png"}, {"type": "image", "uri": "media/ur5fail/35e224b5d2d7a52548d971cf.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0138/ur5fail_exec_success", "group_id": "ur5fail/real_put_banana_in_plate+0/4", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 138, "taskvar": "real_put_banana_in_plate+0", "episode_id": 4, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 271, "role": "outcome"}}
{"state": {"task": "put the scissors and the carrot in the container", "subtask": "grasp scissor", "images": ["start of the subtask, front view", "start of the subtask, left view", "start of the subtask, right view", "end of the subtask, front view", "end of the subtask, left view", "end of the subtask, right view"]}, "questions": {"decision": {"type": "noul", "instructions": "You are observing a manipulation task performed by a UR5 robot arm with a parallel gripper. The state gives the high-level objective, which is divided into subtasks, and the one subtask under evaluation. The attached images show the scene from three fixed cameras (front, left and right views) at the start of that subtask and then at its end, in the order listed under 'images'. Was the subtask performed successfully?", "label": false, "criteria": {"true": "Success: at the end of the subtask the scene shows the subtask accomplished as stated - the right object was grasped, moved, placed or changed, and it ends in the right location or state.", "false": "Failure: the subtask was not accomplished - for example the gripper did not close or missed the object, the wrong object was manipulated, the object ended in the wrong place or state, or nothing meaningful changed."}, "src": "ur5fail_exec_success"}}, "media": [{"type": "image", "uri": "media/ur5fail/851d71871ac2f52df82e11fa.png"}, {"type": "image", "uri": "media/ur5fail/18fe98f12d2ffe7956e3f94a.png"}, {"type": "image", "uri": "media/ur5fail/7103329de79eb58d722a45d8.png"}, {"type": "image", "uri": "media/ur5fail/453f0e02da46926e09b418f0.png"}, {"type": "image", "uri": "media/ur5fail/3f6b315328597534c78bd4e8.png"}, {"type": "image", "uri": "media/ur5fail/4d7944217379d911c883f88c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/exec0139/ur5fail_exec_success", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_exec_success", "label_source": "human_expert", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 139, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "no_gripper_close", "failure_mode_raw": "no_close", "image_sizes": [[256, 256], [256, 256], [256, 256], [256, 256], [256, 256], [256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; start images of all views, then end images"}, "n_media": 6, "approx_tokens": 271, "role": "outcome"}}
{"state": {"task": "move the ball to the right of the banana then move the hammer to the left of the banana, then swap the placement of the hammer and the ball while keeping the ball in the middle", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped ball to the right of the banana"}, {"step": 3, "subtask": "release ball"}, {"step": 4, "subtask": "grasp hammer"}, {"step": 5, "subtask": "move grasped hammer to the left of the banana"}, {"step": 6, "subtask": "release hammer"}, {"step": 7, "subtask": "grasp the ball"}, {"step": 8, "subtask": "move grasped ball to the left of the hammer"}, {"step": 9, "subtask": "release the ball"}, {"step": 10, "subtask": "grasp the hammer"}, {"step": 11, "subtask": "move grasped hammer to the right of the banana"}, {"step": 12, "subtask": "release hammer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/ad4501b9673211a57736fdcf.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0000/ur5fail_plan_correct", "group_id": "ur5fail/real_pick_tool+0/173817592039", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 0, "taskvar": "real_pick_tool+0", "episode_id": 173817592039, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 425, "role": "error_localization"}}
{"state": {"task": "move the ball to the right of the banana then move the hammer to the left of the banana, then swap the placement of the hammer and the ball while keeping the ball in the middle", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped ball to the right of the banana"}, {"step": 3, "subtask": "release ball"}, {"step": 4, "subtask": "grasp hammer"}, {"step": 5, "subtask": "move grasped hammer to the left of the banana"}, {"step": 6, "subtask": "release hammer"}, {"step": 7, "subtask": "move grasped ball to the left of the hammer"}, {"step": 8, "subtask": "grasp the ball"}, {"step": 9, "subtask": "release the ball"}, {"step": 10, "subtask": "grasp the hammer"}, {"step": 11, "subtask": "move grasped hammer to the right of the banana"}, {"step": 12, "subtask": "release hammer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/ad4501b9673211a57736fdcf.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0001/ur5fail_plan_correct", "group_id": "ur5fail/real_pick_tool+0/173817592039", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 1, "taskvar": "real_pick_tool+0", "episode_id": 173817592039, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 425, "role": "error_localization"}}
{"state": {"task": "put the pizza to the upper right part of the table, then to the bottom left part of the table, then put the donut to the upper left part of the table, then to the bottom right part", "plan": [{"step": 1, "subtask": "grasp pizza"}, {"step": 2, "subtask": "move grasped pizza to the upper right part of the table"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pizza"}, {"step": 5, "subtask": "move grasped pizza to the bottom left part of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp donut"}, {"step": 8, "subtask": "move grasped donut to the upper left part of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp donut"}, {"step": 11, "subtask": "move grasped donut to the bottom right part"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0002/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 2, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "place the pizza to the box, then remove it, then put the strawberry in the box, then put the donut next to the corn", "plan": [{"step": 1, "subtask": "grasp pizza"}, {"step": 2, "subtask": "move grasped pizza to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pizza"}, {"step": 5, "subtask": "move grasped pizza out"}, {"step": 6, "subtask": "move grasped pizza to the table"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp donut"}, {"step": 9, "subtask": "move grasped donut next to the corn"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp strawberry"}, {"step": 12, "subtask": "move grasped strawberry to the box"}, {"step": 13, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0003/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 3, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 418, "role": "error_localization"}}
{"state": {"task": "place the pizza to the box, then remove it, then put the strawberry in the box, then put the donut next to the corn", "plan": [{"step": 1, "subtask": "grasp pizza"}, {"step": 2, "subtask": "move grasped pizza to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pizza"}, {"step": 5, "subtask": "move grasped pizza out"}, {"step": 6, "subtask": "move grasped pizza to the table"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp strawberry"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "move grasped strawberry to the box"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp donut"}, {"step": 13, "subtask": "move grasped donut next to the corn"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0004/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 4, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "place the pizza to the box, then remove it, then put the strawberry in the box, then put the donut next to the corn", "plan": [{"step": 1, "subtask": "grasp pizza"}, {"step": 2, "subtask": "move grasped pizza to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "move grasped pizza out"}, {"step": 5, "subtask": "move grasped pizza to the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to the box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp donut"}, {"step": 11, "subtask": "move grasped donut next to the corn"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0005/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 5, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 402, "role": "error_localization"}}
{"state": {"task": "put the strawberry to the upper right part of the table, then to the bottom left part of the table, then to the upper left part of the table, then to the bottom right part", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the upper right part of the table"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped strawberry to the upper left part of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to the bottom left part of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped strawberry to the bottom right part"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0006/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 6, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 424, "role": "error_localization"}}
{"state": {"task": "put the strawberry to the upper right part of the table, then to the bottom left part of the table, then to the upper left part of the table, then to the bottom right part", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the upper right part of the table"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped strawberry to the bottom left part of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to the bottom right part"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0007/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 7, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 369, "role": "error_localization"}}
{"state": {"task": "put the strawberry to the upper right part of the table, then to the bottom left part of the table, then to the upper left part of the table, then to the bottom right part", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the upper right part of the table"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped strawberry to the bottom left part of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to the upper right part of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped strawberry to the bottom right part"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0008/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 8, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 424, "role": "error_localization"}}
{"state": {"task": "put 3 objects into the box", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp donut"}, {"step": 5, "subtask": "move grasped donut to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp corn"}, {"step": 8, "subtask": "release"}, {"step": 9, "subtask": "move grasped corn to box"}, {"step": 10, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0009/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 9, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 339, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box then the donut in the box then the pizza to the right of the corn", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp chicken"}, {"step": 5, "subtask": "move grasped chicken to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pizza"}, {"step": 8, "subtask": "move grasped pizza to the right of the corn"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0010/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 10, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box then the donut in the box then the pizza to the right of the corn", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pizza"}, {"step": 5, "subtask": "move grasped pizza to the right of the corn"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp donut"}, {"step": 8, "subtask": "move grasped donut to box"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0011/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 11, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 344, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "move grasped object to box"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0012/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 12, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 270, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped tuna can to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp corn"}, {"step": 5, "subtask": "move grasped corn to box"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/d9769f6524f9006d490e17d8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0013/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 13, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 286, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped tuna can to box"}, {"step": 3, "subtask": "release grasped tuna can"}, {"step": 4, "subtask": "grasp corn"}, {"step": 5, "subtask": "move grasped corn to box"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0014/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 14, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 290, "role": "error_localization"}}
{"state": {"task": "put the banana then the black grapes in the box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black grapes"}, {"step": 5, "subtask": "move grasped object to box"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0015/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 15, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 281, "role": "error_localization"}}
{"state": {"task": "put the banana then the black grapes in the box", "plan": [{"step": 1, "subtask": "move grasped banana to box"}, {"step": 2, "subtask": "release"}, {"step": 3, "subtask": "grasp black grapes"}, {"step": 4, "subtask": "move grasped object to box"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0016/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 16, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 266, "role": "error_localization"}}
{"state": {"task": "put the banana then the black grapes in the box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black grapes"}, {"step": 5, "subtask": "release"}, {"step": 6, "subtask": "move grasped object to box"}, {"step": 7, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0017/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 17, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 294, "role": "error_localization"}}
{"state": {"task": "put all the fruits you see in the box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black grapes"}, {"step": 5, "subtask": "move grasped black grapes to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp green pear"}, {"step": 11, "subtask": "move grasped green pear to box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp purple plum"}, {"step": 14, "subtask": "move grasped purple plum to box"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0018/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 18, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 429, "role": "error_localization"}}
{"state": {"task": "put all the fruits you see next to the strawberry", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana to strawberry"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black grapes"}, {"step": 5, "subtask": "move grasped black grapes to strawberry"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped green pear to strawberry"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp purple plum"}, {"step": 11, "subtask": "move grasped purple plum to strawberry"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp yellow peach"}, {"step": 14, "subtask": "move grasped yellow peach to strawberry"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0019/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 19, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box, put the banana next to the strawberry in the box, remove the yellow peach from it, put it back", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped banana next to the strawberry in the box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp yellow peach"}, {"step": 8, "subtask": "move grasped yellow peach to table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped yellow peach to the box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0020/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 20, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 404, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box, put the banana next to but not inside the box, remove the yellow peach from it, put it back", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped banana next to but not inside the box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp yellow peach"}, {"step": 8, "subtask": "move grasped yellow peach to table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped yellow peach to the box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0021/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 21, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 404, "role": "error_localization"}}
{"state": {"task": "put the strawberry next to but not inside the box, remove the yellow peach from it, put the green pear inside the box", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry next to but not inside the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow peach"}, {"step": 5, "subtask": "move grasped yellow peach to table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped green pear to the box"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0022/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 22, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "put all the fruits you see next to the strawberry", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana to strawberry"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black grapes"}, {"step": 5, "subtask": "move grasped black grapes to strawberry"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "move grasped green pear to strawberry"}, {"step": 8, "subtask": "release"}, {"step": 9, "subtask": "grasp purple plum"}, {"step": 10, "subtask": "move grasped purple plum to strawberry"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp yellow peach"}, {"step": 13, "subtask": "move grasped yellow peach to strawberry"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0023/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 23, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 415, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped lemon to table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp banana"}, {"step": 12, "subtask": "move grasped banana to box"}, {"step": 13, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0024/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 24, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 404, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana and purple plum in the box, then remove the banana", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped lemon to table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp banana"}, {"step": 12, "subtask": "move grasped banana to box"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp purple plum"}, {"step": 15, "subtask": "move grasped purple plum to box"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp banana"}, {"step": 18, "subtask": "move grasped object out"}, {"step": 19, "subtask": "move grasped banana to table"}, {"step": 20, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0025/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 25, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 530, "role": "error_localization"}}
{"state": {"task": "put the strawberry on the upper right corner of the box, then put the lemon on the upper left corner of the box, then put the yellow peach on the lower right corner of the box, then put the purple plum on the lower left corner of the box", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the upper right corner of the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp lemon"}, {"step": 5, "subtask": "move grasped lemon to the upper left corner of the box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp yellow peach"}, {"step": 8, "subtask": "move grasped yellow peach to the lower right corner of the box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp purple plum"}, {"step": 11, "subtask": "move grasped purple plum to the lower left corner of the box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0026/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 26, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 445, "role": "error_localization"}}
{"state": {"task": "put the strawberry on the upper right corner of the box, then put the lemon on the upper left corner of the box, then remove them both from the box and put the banana instead into it", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the upper right corner of the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp lemon"}, {"step": 5, "subtask": "move grasped lemon to the upper left corner of the box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry out"}, {"step": 9, "subtask": "move grasped strawberry to the table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp lemon"}, {"step": 12, "subtask": "move grasped lemon to the table"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp banana"}, {"step": 15, "subtask": "move grasped banana to the box"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0027/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 27, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 486, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped lemon to table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp banana"}, {"step": 12, "subtask": "move grasped banana to box"}, {"step": 13, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/8e2760d2c3383c0c8ec46633.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0028/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 28, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 404, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "move grasped object out"}, {"step": 8, "subtask": "move grasped lemon to table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped banana to box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0029/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 29, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 388, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp peach"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped lemon to table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp banana"}, {"step": 12, "subtask": "move grasped banana to box"}, {"step": 13, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0030/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 30, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 404, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped lemon to table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp banana"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "move grasped banana to box"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0031/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 31, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 418, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped object to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "move grasped object to box"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0032/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 32, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 264, "role": "error_localization"}}
{"state": {"task": "put all the fruits in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped strawberry to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped banana to box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped yellow peach to box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp purple plum"}, {"step": 14, "subtask": "move grasped purple plum to box"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0033/ur5fail_plan_correct", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 33, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 426, "role": "error_localization"}}
{"state": {"task": "put the ball in the top part of the drawer", "plan": [{"step": 1, "subtask": "grasp top part of the drawer"}, {"step": 2, "subtask": "move grasped object out"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to top part of the drawer"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0034/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 34, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 287, "role": "error_localization"}}
{"state": {"task": "remove from the middle drawer the screwdriver and put it on the table, then put the frog toy in the middle drawer, then close it, then remove from the top drawer the hammer and put it on the table, and put the can in the top drawer, then close it", "plan": [{"step": 1, "subtask": "grasp middle drawer handle"}, {"step": 2, "subtask": "move grasped object to drawer open position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp screwdriver"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to the table"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp frog toy"}, {"step": 9, "subtask": "move grasped object to middle drawer"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp middle drawer handle"}, {"step": 12, "subtask": "move grasped object to drawer close position"}, {"step": 13, "subtask": "grasp top drawer handle"}, {"step": 14, "subtask": "move grasped object to drawer open position"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp hammer"}, {"step": 17, "subtask": "move grasped object out"}, {"step": 18, "subtask": "move grasped object to the table"}, {"step": 19, "subtask": "release"}, {"step": 20, "subtask": "grasp can"}, {"step": 21, "subtask": "move grasped object to top drawer"}, {"step": 22, "subtask": "release"}, {"step": 23, "subtask": "grasp top drawer handle"}, {"step": 24, "subtask": "move grasped object to drawer close position"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0035/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 35, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 647, "role": "error_localization"}}
{"state": {"task": "put the frog toy, the can, the ball, and the pink cube into the top drawer", "plan": [{"step": 1, "subtask": "grasp frog toy"}, {"step": 2, "subtask": "move grasped object to top drawer"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped object to top drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp ball"}, {"step": 8, "subtask": "move grasped object to top drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pink cube"}, {"step": 11, "subtask": "move grasped object to top drawer"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/21cbd33925e9914daa49d862.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0036/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 36, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 390, "role": "error_localization"}}
{"state": {"task": "put the frog toy, the can, the ball, and the pink cube into the top drawer, and close it", "plan": [{"step": 1, "subtask": "grasp frog toy"}, {"step": 2, "subtask": "move grasped object to top drawer"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp can"}, {"step": 5, "subtask": "move grasped object to top drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp ball"}, {"step": 8, "subtask": "move grasped object to top drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pink cube"}, {"step": 11, "subtask": "move grasped object to top drawer"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp top drawer handle"}, {"step": 14, "subtask": "move grasped object to drawer close position"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/21cbd33925e9914daa49d862.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0037/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 37, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 447, "role": "error_localization"}}
{"state": {"task": "remove the screwdriver, and hammer from the bottom drawer to put them on the table, then close the drawer", "plan": [{"step": 1, "subtask": "grasp bottom drawer handle"}, {"step": 2, "subtask": "move grasped object to drawer open position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp screwdriver"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to table"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp hammer"}, {"step": 9, "subtask": "move grasped object out"}, {"step": 10, "subtask": "move grasped object to table"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp bottom drawer handle"}, {"step": 13, "subtask": "move grasped object to drawer close position"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0038/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 38, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "remove the screwdriver, and hammer from the bottom drawer to put them on the table, then put instead the pink cube and can in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp bottom drawer handle"}, {"step": 2, "subtask": "move grasped object bottom drawer open position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp screwdriver"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to table"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp hammer"}, {"step": 9, "subtask": "move grasped object out"}, {"step": 10, "subtask": "move grasped object to table"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp pink cube"}, {"step": 13, "subtask": "move grasped object to bottom drawer"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "grasp can"}, {"step": 16, "subtask": "move grasped object to bottom drawer"}, {"step": 17, "subtask": "release"}, {"step": 18, "subtask": "grasp bottom drawer handle"}, {"step": 19, "subtask": "move grasped object bottom drawer close position"}, {"step": 20, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0039/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 39, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 546, "role": "error_localization"}}
{"state": {"task": "put the ball in the top drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped object to top drawer"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp handle of the top drawer"}, {"step": 5, "subtask": "move grasped object to closed drawer position"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0040/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 40, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 289, "role": "error_localization"}}
{"state": {"task": "put the ball in the top drawer", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped object to top drawer"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0041/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 41, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 232, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp handle of the bottom drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cube"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp frog toy"}, {"step": 11, "subtask": "move grasped object to bottom drawer"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp handle of the bottom drawer"}, {"step": 14, "subtask": "move grasped object to closed drawer position"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0042/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 42, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 449, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped object to bottom drawer"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pink cube"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp frog toy"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp handle of the bottom drawer"}, {"step": 11, "subtask": "move grasped object to closed drawer position"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0043/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 43, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 394, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp handle of the bottom drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cube"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp frog toy"}, {"step": 11, "subtask": "move grasped object to top drawer"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp handle of the bottom drawer"}, {"step": 14, "subtask": "move grasped object to closed drawer position"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0044/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 44, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 449, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp handle of the bottom drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cube"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp frog toy"}, {"step": 11, "subtask": "move grasped object to bottom drawer"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp handle of the middle drawer"}, {"step": 14, "subtask": "move grasped object to closed drawer position"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0045/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 45, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 449, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp handle of the bottom drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cube"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "move grasped object to bottom drawer"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp handle of the bottom drawer"}, {"step": 13, "subtask": "move grasped object to closed drawer position"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0046/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 46, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "put the ball, then frog toy, then pink cube in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp handle of the bottom drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp frog toy"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pink cube"}, {"step": 11, "subtask": "move grasped object to bottom drawer"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp handle of the bottom drawer"}, {"step": 14, "subtask": "move grasped object to closed drawer position"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0047/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 47, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 451, "role": "error_localization"}}
{"state": {"task": "put all the objects you see in the middle drawer", "plan": [{"step": 1, "subtask": "grasp handle of the middle drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to middle drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp frog toy"}, {"step": 8, "subtask": "move grasped object to middle drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pink cube"}, {"step": 11, "subtask": "move grasped object to middle drawer"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp can"}, {"step": 14, "subtask": "move grasped object to middle drawer"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0048/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 48, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 436, "role": "error_localization"}}
{"state": {"task": "put ball in top drawer, close it, then put pink cube in middle drawer and close it", "plan": [{"step": 1, "subtask": "grasp handle of the top drawer"}, {"step": 2, "subtask": "move grasped object out"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to top drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp handle of the top drawer"}, {"step": 8, "subtask": "move grasped object to closed drawer position"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp handle of the middle drawer"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "move grasped object to open drawer position"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp pink cube"}, {"step": 15, "subtask": "move grasped object to middle drawer"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp handle of the middle drawer"}, {"step": 18, "subtask": "move grasped object to closed drawer position"}, {"step": 19, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0049/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 49, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 520, "role": "error_localization"}}
{"state": {"task": "put ball in top drawer, close it, then put pink cube in middle drawer and close it", "plan": [{"step": 1, "subtask": "grasp handle of the top drawer"}, {"step": 2, "subtask": "move grasped object out"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to top drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp handle of the top drawer"}, {"step": 8, "subtask": "move grasped object to closed drawer position"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp handle of the middle drawer"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "move grasped object to open drawer position"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp pink cube"}, {"step": 15, "subtask": "move grasped object to middle drawer"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp handle of the top drawer"}, {"step": 18, "subtask": "move grasped object to closed drawer position"}, {"step": 19, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0050/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 50, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 520, "role": "error_localization"}}
{"state": {"task": "put the ball in the top part of the drawer", "plan": [{"step": 1, "subtask": "grasp top part of the drawer"}, {"step": 2, "subtask": "release"}, {"step": 3, "subtask": "grasp ball"}, {"step": 4, "subtask": "move grasped object to top part of the drawer"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0051/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 51, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 270, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the cyan plate, then remove it back to the table, then put the strawberry in the plate, then remove it back to the table, finally put them both back into the plate", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to cyan plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tuna can"}, {"step": 5, "subtask": "move grasped object to table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped object to cyan plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped object to table"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp tuna can"}, {"step": 14, "subtask": "move grasped object to cyan plate"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp strawberry"}, {"step": 17, "subtask": "move grasped object to cyan plate"}, {"step": 18, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/23196e5cafe425f2d66448cd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0052/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 52, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 511, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the plate, then the tuna can on the upper right of the table, then the tuna can on the lower right of the table, then remove the strawberry from the plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tuna can"}, {"step": 5, "subtask": "move grasped object to upper right of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped object to lower right of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped object to table"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/23196e5cafe425f2d66448cd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0053/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 53, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 412, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the plate, then the tuna can on the upper right of the table, then the tuna can on the lower right of the table, then remove the strawberry from the plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tuna can"}, {"step": 5, "subtask": "move grasped object to upper right of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp tuna can"}, {"step": 8, "subtask": "move grasped object to lower right of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp tuna can"}, {"step": 11, "subtask": "move grasped object to table"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/23196e5cafe425f2d66448cd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0054/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 54, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 414, "role": "error_localization"}}
{"state": {"task": "grasp the strawberry and place it in the blue plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to blue plate"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/184cbf96c4115286374071b8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0055/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816492337", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 55, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816492337, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 236, "role": "error_localization"}}
{"state": {"task": "place every object you see in the blue plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to blue plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp hot dog"}, {"step": 5, "subtask": "move grasped object to blue plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp bottle"}, {"step": 8, "subtask": "move grasped object to blue plate"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/184cbf96c4115286374071b8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0056/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816492337", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 56, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816492337, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 329, "role": "error_localization"}}
{"state": {"task": "place every object you see in the blue plate, then remove them all from the plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to blue plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp hot dog"}, {"step": 5, "subtask": "move grasped object to blue plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp bottle"}, {"step": 8, "subtask": "move grasped object to blue plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped object to table"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp hot dog"}, {"step": 14, "subtask": "move grasped object to table"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp bottle"}, {"step": 17, "subtask": "move grasped object to table"}, {"step": 18, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/184cbf96c4115286374071b8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0057/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816492337", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 57, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816492337, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 485, "role": "error_localization"}}
{"state": {"task": "place eatable food in the blue plate, place drink next to blue plate but not inside plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to blue plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp hot dog"}, {"step": 5, "subtask": "move grasped object to blue plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp bottle"}, {"step": 8, "subtask": "move grasped object next to blue plate"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/184cbf96c4115286374071b8.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0058/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816492337", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 58, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816492337, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 340, "role": "error_localization"}}
{"state": {"task": "place eatable food in the blue plate, then remove them", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object to blue plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp hot dog"}, {"step": 5, "subtask": "move grasped object to blue plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped object to table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp hot dog"}, {"step": 11, "subtask": "move grasped object to table"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/66f558ab24ef777dc5829b87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0059/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816518206", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 59, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816518206, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 383, "role": "error_localization"}}
{"state": {"task": "place all the fruits you see in the blue plate", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object to blue plate"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/66f558ab24ef777dc5829b87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0060/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816518206", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 60, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816518206, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 236, "role": "error_localization"}}
{"state": {"task": "place all the fruits you see in the blue plate", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped green pear to blue plate"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/aa7b95d5f56ae99a9ba77c2b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0061/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816518206", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 61, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816518206, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "pick the green pear and place it next to the ball", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object to container"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/66f558ab24ef777dc5829b87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0062/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816518206", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 62, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816518206, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 236, "role": "error_localization"}}
{"state": {"task": "pick the green pear and place it next to box", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object into the wooden box box"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/799fe8e2dfc95526aa62ae73.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0063/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816539090", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 63, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816539090, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 238, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the wooden box", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped green pear to the wooden box"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/46a025e974f6b82cd32eb267.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0064/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816550049", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 64, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816550049, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "put into the box all the objects that are on the table, but before that, if you see any object inside the box, remove it", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp hot dog"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp chips container"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/346b0c0621b7ca68f0b77d98.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0065/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816553691", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 65, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816553691, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 350, "role": "error_localization"}}
{"state": {"task": "put into the box all the objects that are on the table, but before that, if you see any object inside the box, remove it", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object out"}, {"step": 3, "subtask": "move grasped object to the table"}, {"step": 4, "subtask": "release"}, {"step": 5, "subtask": "grasp hot dog"}, {"step": 6, "subtask": "move grasped object to wooden box"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp chips container"}, {"step": 9, "subtask": "move grasped object to wooden box"}, {"step": 10, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/346b0c0621b7ca68f0b77d98.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0066/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816553691", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 66, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816553691, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 368, "role": "error_localization"}}
{"state": {"task": "place the healthiest food you see in the wooden box", "plan": [{"step": 1, "subtask": "grasp chocolate"}, {"step": 2, "subtask": "move grasped object to the wooden box"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/98405d9e80f43cbc72e8a360.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0067/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816578860", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 67, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816578860, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "grab the healthiest food", "plan": [{"step": 1, "subtask": "grasp carrot"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/a721c33537bfdecba849aa6d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0068/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817342016", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 68, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817342016, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 198, "role": "error_localization"}}
{"state": {"task": "flip upright the bottle, then put the bread to the left of the bottle, then put the bread to the right of the bottle, then on top of the pop corn box", "plan": [{"step": 1, "subtask": "grasp bottle"}, {"step": 2, "subtask": "rotate grasped object"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp bread"}, {"step": 5, "subtask": "move grasped object to the right of the bottle"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp bread"}, {"step": 8, "subtask": "move grasped object to the left of the bottle"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped object to the top of pop corn box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/a721c33537bfdecba849aa6d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0069/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817342016", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 69, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817342016, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 411, "role": "error_localization"}}
{"state": {"task": "pick up the healthiest food, then move it to each object of the table sequentially, finally put it next to the coke bottle", "plan": [{"step": 1, "subtask": "grasp carrot"}, {"step": 2, "subtask": "move grasped object to the coke bottle"}, {"step": 3, "subtask": "move grasped object to the bread"}, {"step": 4, "subtask": "move grasped object next to the coke bottle"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/138f5eaf01013f703f20adfe.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0070/ur5fail_plan_correct", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817351532", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 70, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817351532, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 292, "role": "error_localization"}}
{"state": {"task": "stack the blue cup on top of the red cube, then put the ball inside the blue cup, then unstack the tower without making anything fall", "plan": [{"step": 1, "subtask": "grasp blue cup"}, {"step": 2, "subtask": "move grasped object to top of the red cube"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object into the blue cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp ball"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped object to the table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp red cube"}, {"step": 12, "subtask": "move grasped object to the table"}, {"step": 13, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/458ccd670843141efd784cbd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0071/ur5fail_plan_correct", "group_id": "ur5fail/real_sport+0/173817680775", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 71, "taskvar": "real_sport+0", "episode_id": 173817680775, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 422, "role": "error_localization"}}
{"state": {"task": "move next to the carrot every object you see", "plan": [{"step": 1, "subtask": "grasp blue cup"}, {"step": 2, "subtask": "move grasped object to next to the carrot"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to next to the carrot"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp red cube"}, {"step": 8, "subtask": "move grasped object to next to the carrot"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/458ccd670843141efd784cbd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0072/ur5fail_plan_correct", "group_id": "ur5fail/real_sport+0/173817680775", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 72, "taskvar": "real_sport+0", "episode_id": 173817680775, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 336, "role": "error_localization"}}
{"state": {"task": "move next to the carrot every object you see", "plan": [{"step": 1, "subtask": "grasp blue cup"}, {"step": 2, "subtask": "move grasped object to next to the carrot"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to next to the carrot"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp carrot"}, {"step": 8, "subtask": "move grasped object to next to the carrot"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/458ccd670843141efd784cbd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0073/ur5fail_plan_correct", "group_id": "ur5fail/real_sport+0/173817680775", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 73, "taskvar": "real_sport+0", "episode_id": 173817680775, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 335, "role": "error_localization"}}
{"state": {"task": "stack a pink cup onto a black cup onto a cyan cup onto a yellow cup onto a navy cup", "plan": [{"step": 1, "subtask": "grasp pink cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black cup"}, {"step": 5, "subtask": "move grasped object to top of cyan cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp cyan cup"}, {"step": 8, "subtask": "move grasped object to top of navy cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp navy cup"}, {"step": 11, "subtask": "move grasped object to top of yellow cup"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0074/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 74, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 401, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of navy cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp cyan cup"}, {"step": 5, "subtask": "move grasped object to top of yellow cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp black cup"}, {"step": 8, "subtask": "move grasped object to top of cyan cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0075/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 75, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 341, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of navy cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black cup"}, {"step": 5, "subtask": "move grasped object to top of cyan cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cup"}, {"step": 8, "subtask": "move grasped object to top of black cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0076/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 76, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 341, "role": "error_localization"}}
{"state": {"task": "create a stack of four cups: yellow on navy on pink on black cups", "plan": [{"step": 1, "subtask": "grasp pink cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp navy cup"}, {"step": 5, "subtask": "move grasped object to top of yellow cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp yellow cup"}, {"step": 8, "subtask": "move grasped object to top of pink cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0077/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 77, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 343, "role": "error_localization"}}
{"state": {"task": "create two stacks of cups: one with the yellow and navy, then one with the cyan, black and pink cups", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of navy cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black cup"}, {"step": 5, "subtask": "move grasped object to top of pink cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp cyan cup"}, {"step": 8, "subtask": "move grasped object to top of navy cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0078/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 78, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 351, "role": "error_localization"}}
{"state": {"task": "create two stacks of cups: one with the yellow and navy, then one with the cyan, and pink cups", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "release"}, {"step": 3, "subtask": "move grasped object to top of navy cup"}, {"step": 4, "subtask": "release"}, {"step": 5, "subtask": "grasp cyan cup"}, {"step": 6, "subtask": "move grasped object to top of pink cup"}, {"step": 7, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0079/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 79, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 313, "role": "error_localization"}}
{"state": {"task": "create a stack of cups: navy on top of pink, pink on top of cyan, cyan on top of yellow, yellow on top of black", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp cyan cup"}, {"step": 5, "subtask": "move grasped object to top of yellow cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cup"}, {"step": 8, "subtask": "move grasped object to top of cyan cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp navy cup"}, {"step": 11, "subtask": "move grasped object to top of pink cup"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0080/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 80, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 410, "role": "error_localization"}}
{"state": {"task": "create a stack of cups: navy on top of pink, pink on top of cyan, cyan on top of yellow, yellow on top of black", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp cyan cup"}, {"step": 5, "subtask": "move grasped object to top of pink cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cup"}, {"step": 8, "subtask": "move grasped object to top of cyan cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp navy cup"}, {"step": 11, "subtask": "move grasped object to top of pink cup"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0081/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 81, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 410, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table", "plan": [{"step": 1, "subtask": "grasp black cup"}, {"step": 2, "subtask": "move grasped object to top of red cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of black cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of yellow cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp cyan cup"}, {"step": 11, "subtask": "move grasped object to top of green cup"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0082/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 82, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 394, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table", "plan": [{"step": 1, "subtask": "grasp black cup"}, {"step": 2, "subtask": "move grasped object to top of red cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of black cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of yellow cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0083/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 83, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 341, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table, except the cyan cup that should be moved next but not on top of the stack", "plan": [{"step": 1, "subtask": "grasp black cup"}, {"step": 2, "subtask": "move grasped object to top of red cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of black cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of yellow cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp cyan cup"}, {"step": 11, "subtask": "move grasped object next to the stack"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0084/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 84, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 410, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table, except the cyan cup that should be moved next but not on top of the stack", "plan": [{"step": 1, "subtask": "grasp black cup"}, {"step": 2, "subtask": "move grasped object to top of red cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of green cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of yellow cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp cyan cup"}, {"step": 11, "subtask": "move grasped object next to the stack"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0085/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 85, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 410, "role": "error_localization"}}
{"state": {"task": "stack a black cup onto a red cup", "plan": [{"step": 1, "subtask": "grasp red cup"}, {"step": 2, "subtask": "move grasped object to black cup"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0086/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 86, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 234, "role": "error_localization"}}
{"state": {"task": "put the red cup on top of black cup, then put the yellow cup on top of cyan cup, then put the green cup on top of the red cup, then unstack the stack containing the red cup", "plan": [{"step": 1, "subtask": "grasp red cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of cyan cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of cyan cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp green cup"}, {"step": 11, "subtask": "move grasped object out"}, {"step": 12, "subtask": "move grasped object to table"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp red cup"}, {"step": 15, "subtask": "move grasped object out"}, {"step": 16, "subtask": "move grasped object to table"}, {"step": 17, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0087/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 87, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 506, "role": "error_localization"}}
{"state": {"task": "stack a stack of cups with the black cup as the base, next the red cup, then the yellow cup, then the green cup", "plan": [{"step": 1, "subtask": "grasp red cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of red cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of yellow cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0088/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 88, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 355, "role": "error_localization"}}
{"state": {"task": "stack a stack of cups with the black cup as the base, next the red cup, then the yellow cup, then the green cup, then unstack them", "plan": [{"step": 1, "subtask": "grasp red cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of red cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of yellow cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp green cup"}, {"step": 11, "subtask": "move grasped object out"}, {"step": 12, "subtask": "move grasped object to table"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp yellow cup"}, {"step": 15, "subtask": "move grasped object out"}, {"step": 16, "subtask": "move grasped object to table"}, {"step": 17, "subtask": "release"}, {"step": 18, "subtask": "grasp red cup"}, {"step": 19, "subtask": "move grasped object out"}, {"step": 20, "subtask": "move grasped object to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0089/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 89, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 564, "role": "error_localization"}}
{"state": {"task": "stack black cup on orange cup", "plan": [{"step": 1, "subtask": "grasp black cup"}, {"step": 2, "subtask": "move grasped black cup on top of orange cup"}, {"step": 3, "subtask": "release black cup"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/c24d8e97cccbe22ab223bb25.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0090/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+5/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 90, "taskvar": "real_stack_cup+5", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 237, "role": "error_localization"}}
{"state": {"task": "stack black cup on orange cup, then unstack them, then restack them, then unstack them again", "plan": [{"step": 1, "subtask": "grasp black cup"}, {"step": 2, "subtask": "move grasped object to top of orange cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black cup"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to table"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp black cup"}, {"step": 9, "subtask": "move grasped object to top of orange cup"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp black cup"}, {"step": 12, "subtask": "move grasped object out"}, {"step": 13, "subtask": "move grasped object to table"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/c24d8e97cccbe22ab223bb25.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0091/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+5/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 91, "taskvar": "real_stack_cup+5", "episode_id": 0, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 433, "role": "error_localization"}}
{"state": {"task": "grasp the cup that is the closest to the red apple, and put it on the pink cup, then unstack them", "plan": [{"step": 1, "subtask": "grasp navy cup"}, {"step": 2, "subtask": "move grasped object to the pink cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp navy cup"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to table"}, {"step": 7, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/6af2e641d831d29db4c016c2.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0092/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup_spatial+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 92, "taskvar": "real_stack_cup_spatial+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 316, "role": "error_localization"}}
{"state": {"task": "grasp the cup that is the closest to the red apple, and put it on the pink cup, then put the apple on top of the stack, then unstack the entire stack", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to the pink cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "release"}, {"step": 6, "subtask": "move grasped object to top of yellow cup"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp red apple"}, {"step": 9, "subtask": "move grasped object out"}, {"step": 10, "subtask": "move grasped object to table"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp yellow cup"}, {"step": 13, "subtask": "move grasped object out"}, {"step": 14, "subtask": "move grasped object to table"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/6af2e641d831d29db4c016c2.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0093/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup_spatial+0/0", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 93, "taskvar": "real_stack_cup_spatial+0", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 461, "role": "error_localization"}}
{"state": {"task": "put into the wooden box every object you see on the table", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp corn"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp red apple"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp green pear"}, {"step": 14, "subtask": "move grasped object to wooden box"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp yellow peach"}, {"step": 17, "subtask": "move grasped object to wooden box"}, {"step": 18, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/1fdde352d142659373f2c9e0.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0094/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 94, "taskvar": "real_put_food_in_box+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 485, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the strawberry, the corn, the green pear, then remove them from the wooden box", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp corn"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped object out"}, {"step": 12, "subtask": "move grasped object to table"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp corn"}, {"step": 15, "subtask": "move grasped object out"}, {"step": 16, "subtask": "move grasped object to table"}, {"step": 17, "subtask": "release"}, {"step": 18, "subtask": "grasp green pear"}, {"step": 19, "subtask": "move grasped object out"}, {"step": 20, "subtask": "move grasped object to table"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/1fdde352d142659373f2c9e0.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0095/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 95, "taskvar": "real_put_food_in_box+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 544, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the tuna can and all the red fruits you see on the table", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp red apple"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/1fdde352d142659373f2c9e0.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0096/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 96, "taskvar": "real_put_food_in_box+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 339, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the strawberry then the banana, then the lemon then the yellow peach", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e84eaff8c1b974397c2a3d23.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0097/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 97, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 388, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the strawberry then the banana, then the lemon then the yellow peach, then put them back on the table", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp yellow peach"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp lemon"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp strawberry"}, {"step": 14, "subtask": "move grasped object out"}, {"step": 15, "subtask": "move grasped object to table"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp banana"}, {"step": 18, "subtask": "move grasped object out"}, {"step": 19, "subtask": "move grasped object to table"}, {"step": 20, "subtask": "release"}, {"step": 21, "subtask": "grasp lemon"}, {"step": 22, "subtask": "move grasped object out"}, {"step": 23, "subtask": "move grasped object to table"}, {"step": 24, "subtask": "release"}, {"step": 25, "subtask": "grasp yellow peach"}, {"step": 26, "subtask": "move grasped object out"}, {"step": 27, "subtask": "move grasped object to table"}, {"step": 28, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e84eaff8c1b974397c2a3d23.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0098/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 98, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 665, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the strawberry then the banana, then the lemon then the yellow peach, then put them back on the table", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow corn"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp strawberry"}, {"step": 14, "subtask": "move grasped object out"}, {"step": 15, "subtask": "move grasped object to table"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp banana"}, {"step": 18, "subtask": "move grasped object out"}, {"step": 19, "subtask": "move grasped object to table"}, {"step": 20, "subtask": "release"}, {"step": 21, "subtask": "grasp lemon"}, {"step": 22, "subtask": "move grasped object out"}, {"step": 23, "subtask": "move grasped object to table"}, {"step": 24, "subtask": "release"}, {"step": 25, "subtask": "grasp yellow peach"}, {"step": 26, "subtask": "move grasped object out"}, {"step": 27, "subtask": "move grasped object to table"}, {"step": 28, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e84eaff8c1b974397c2a3d23.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0099/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 99, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 665, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the strawberry then the banana, then the lemon then the yellow peach, then put them back on the table", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp strawberry"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "move grasped object out"}, {"step": 16, "subtask": "move grasped object to table"}, {"step": 17, "subtask": "release"}, {"step": 18, "subtask": "grasp banana"}, {"step": 19, "subtask": "move grasped object out"}, {"step": 20, "subtask": "move grasped object to table"}, {"step": 21, "subtask": "release"}, {"step": 22, "subtask": "grasp lemon"}, {"step": 23, "subtask": "move grasped object out"}, {"step": 24, "subtask": "move grasped object to table"}, {"step": 25, "subtask": "release"}, {"step": 26, "subtask": "grasp yellow peach"}, {"step": 27, "subtask": "move grasped object out"}, {"step": 28, "subtask": "move grasped object to table"}, {"step": 29, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e84eaff8c1b974397c2a3d23.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0100/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 100, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 679, "role": "error_localization"}}
{"state": {"task": "put the banana in the wooden box then put the strawberry in the wooden box, then remove them both from the wooden box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped object to table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp strawberry"}, {"step": 12, "subtask": "move grasped object out"}, {"step": 13, "subtask": "move grasped object to table"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e84eaff8c1b974397c2a3d23.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0101/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 101, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 427, "role": "error_localization"}}
{"state": {"task": "put into the wooden box all the fruits you see on the table", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp lemon"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/353828c4e2b2f166f915f39f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0102/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/4", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 102, "taskvar": "real_put_food_in_box+0", "episode_id": 4, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 383, "role": "error_localization"}}
{"state": {"task": "remove everything from the wooden box and put them on the table", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object out"}, {"step": 3, "subtask": "move grasped object to table"}, {"step": 4, "subtask": "release"}, {"step": 5, "subtask": "grasp purple plum"}, {"step": 6, "subtask": "move grasped object out"}, {"step": 7, "subtask": "move grasped object to table"}, {"step": 8, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/353828c4e2b2f166f915f39f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0103/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/4", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 103, "taskvar": "real_put_food_in_box+0", "episode_id": 4, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 318, "role": "error_localization"}}
{"state": {"task": "put into the wooden box all the objects you see on the table", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp yellow corn cob"}, {"step": 14, "subtask": "move grasped object to wooden box"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp strawberry"}, {"step": 17, "subtask": "move grasped object to wooden box"}, {"step": 18, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/bd0f90f8f9fb55c990e0d017.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0104/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 104, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 488, "role": "error_localization"}}
{"state": {"task": "put into the wooden box all the objects you see on the table", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp yellow corn cob"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "move grasped object to wooden box"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp strawberry"}, {"step": 18, "subtask": "move grasped object to wooden box"}, {"step": 19, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/bd0f90f8f9fb55c990e0d017.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0105/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 105, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 502, "role": "error_localization"}}
{"state": {"task": "put into the wooden box all the objects you see on the table, then put them all back on the table", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp yellow corn cob"}, {"step": 14, "subtask": "move grasped object to wooden box"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp strawberry"}, {"step": 17, "subtask": "move grasped object to wooden box"}, {"step": 18, "subtask": "release"}, {"step": 19, "subtask": "grasp tuna can"}, {"step": 20, "subtask": "move grasped object out"}, {"step": 21, "subtask": "move grasped object to table"}, {"step": 22, "subtask": "release"}, {"step": 23, "subtask": "grasp red apple"}, {"step": 24, "subtask": "move grasped object out"}, {"step": 25, "subtask": "move grasped object to table"}, {"step": 26, "subtask": "release"}, {"step": 27, "subtask": "grasp green pear"}, {"step": 28, "subtask": "move grasped object out"}, {"step": 29, "subtask": "move grasped object to table"}, {"step": 30, "subtask": "release"}, {"step": 31, "subtask": "grasp yellow peach"}, {"step": 32, "subtask": "move grasped object to wooden box"}, {"step": 33, "subtask": "release"}, {"step": 34, "subtask": "grasp yellow corn cob"}, {"step": 35, "subtask": "move grasped object out"}, {"step": 36, "subtask": "move grasped object to table"}, {"step": 37, "subtask": "release"}, {"step": 38, "subtask": "grasp strawberry"}, {"step": 39, "subtask": "move grasped object out"}, {"step": 40, "subtask": "move grasped object to table"}, {"step": 41, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/bd0f90f8f9fb55c990e0d017.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0106/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 106, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 888, "role": "error_localization"}}
{"state": {"task": "put the strawberry to the right of red apple, then put the green pear to the left of the yellow peach, then put the tuna can into the wooden box", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to the right of red apple"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp green pear"}, {"step": 5, "subtask": "move grasped object to the left of yellow peach"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp tuna can"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/bd0f90f8f9fb55c990e0d017.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0107/ur5fail_plan_correct", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 107, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 359, "role": "error_localization"}}
{"state": {"task": "put into the black box the tool you see on the table, then put it back where it was", "plan": [{"step": 1, "subtask": "grasp screwdriver"}, {"step": 2, "subtask": "move grasped object to black box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp screwdriver"}, {"step": 5, "subtask": "move grasped object to previous screwdriver position"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/73b55c9c94813b749a9ee01f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0108/ur5fail_plan_correct", "group_id": "ur5fail/real_pick_tool_gondola+0/1", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 108, "taskvar": "real_pick_tool_gondola+0", "episode_id": 1, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 296, "role": "error_localization"}}
{"state": {"task": "clean the table by storing everything in the cardboard box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp green plate"}, {"step": 5, "subtask": "move grasped object to cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp white baseball ball"}, {"step": 8, "subtask": "move grasped object to cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp scissors"}, {"step": 11, "subtask": "move grasped object to cardboard box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pen"}, {"step": 14, "subtask": "move grasped object to cardboard box"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/47ff4890929bdb729765135b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0109/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/1", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 109, "taskvar": "real_put_carrot_in_container+0", "episode_id": 1, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 432, "role": "error_localization"}}
{"state": {"task": "clean the table by storing everything in the cardboard box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp green plate"}, {"step": 5, "subtask": "move grasped object to cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp white baseball ball"}, {"step": 8, "subtask": "move grasped object to cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to cardboard box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/47ff4890929bdb729765135b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0110/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/1", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 110, "taskvar": "real_put_carrot_in_container+0", "episode_id": 1, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 382, "role": "error_localization"}}
{"state": {"task": "clean the table by storing everything in the cardboard box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp green plate"}, {"step": 5, "subtask": "move grasped object to cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp white baseball ball"}, {"step": 8, "subtask": "move grasped object to cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp scissors"}, {"step": 11, "subtask": "move grasped object to cardboard box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pen"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "move grasped object to cardboard box"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/47ff4890929bdb729765135b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0111/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/1", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 111, "taskvar": "real_put_carrot_in_container+0", "episode_id": 1, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 446, "role": "error_localization"}}
{"state": {"task": "put everything in the big cardboard box, except the wooden box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to big cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp white baseball ball"}, {"step": 5, "subtask": "move grasped object to big cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to big cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to big cardboard box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0112/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 112, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "put scissors and white baseball ball in the big cardboard box, then put the banana and the pen in the wooden box", "plan": [{"step": 1, "subtask": "grasp scissors"}, {"step": 2, "subtask": "move grasped object to big cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp white baseball ball"}, {"step": 5, "subtask": "move grasped object to big cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0113/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 113, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 396, "role": "error_localization"}}
{"state": {"task": "put into the big cardboard box the banana, then the white baseball ball, then the scissors, then the pen, and finally put them all into the wooden box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to big cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp white baseball ball"}, {"step": 5, "subtask": "move grasped object to big cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to big cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp white baseball ball"}, {"step": 14, "subtask": "move grasped object to wooden box"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp scissors"}, {"step": 17, "subtask": "move grasped object to wooden box"}, {"step": 18, "subtask": "release"}, {"step": 19, "subtask": "grasp pen"}, {"step": 20, "subtask": "move grasped object to wooden box"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0114/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 114, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 558, "role": "error_localization"}}
{"state": {"task": "put into the big cardboard box the banana, then the white baseball ball, then the scissors, then the pen, and finally put them all into the wooden box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to big cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp white baseball ball"}, {"step": 5, "subtask": "move grasped object to big cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to big cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to big cardboard box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp banana"}, {"step": 14, "subtask": "move grasped object to wooden box"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp white baseball ball"}, {"step": 17, "subtask": "move grasped object to wooden box"}, {"step": 18, "subtask": "release"}, {"step": 19, "subtask": "move grasped object to wooden box"}, {"step": 20, "subtask": "grasp scissors"}, {"step": 21, "subtask": "release"}, {"step": 22, "subtask": "grasp pen"}, {"step": 23, "subtask": "move grasped object to wooden box"}, {"step": 24, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0115/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 115, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 609, "role": "error_localization"}}
{"state": {"task": "put into the big cardboard box the banana, then the white baseball ball, then the scissors, then the pen, and finally put the pen and scissors into the wooden box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to big cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp white baseball ball"}, {"step": 5, "subtask": "move grasped object to big cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to big cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to big cardboard box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pen"}, {"step": 14, "subtask": "move grasped object out"}, {"step": 15, "subtask": "move grasped object to wooden box"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp scissors"}, {"step": 18, "subtask": "move grasped object out"}, {"step": 19, "subtask": "move grasped object to wooden box"}, {"step": 20, "subtask": "release"}, {"step": 21, "subtask": "grasp banana"}, {"step": 22, "subtask": "move grasped object out"}, {"step": 23, "subtask": "move grasped object to wooden box"}, {"step": 24, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0116/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 116, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 613, "role": "error_localization"}}
{"state": {"task": "put every object in the largest box (except the other box)", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to wooden square box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to wooden square box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to wooden square box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to wooden square box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0117/ur5fail_plan_correct", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 117, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 385, "role": "error_localization"}}
{"state": {"task": "stack the pink cup onto the black cup then the black cup onto the cyan cup, finally stack the cyan cup onto the yellow cup", "plan": [{"step": 1, "subtask": "grasp pink cup"}, {"step": 2, "subtask": "move grasped object to black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black cup"}, {"step": 5, "subtask": "move grasped object to cyan cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp cyan cup"}, {"step": 8, "subtask": "move grasped object to yellow cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e15a2ee7278ed7615582f75b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0118/ur5fail_plan_correct", "group_id": "ur5fail/real_stack_cup+3/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 118, "taskvar": "real_stack_cup+3", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 348, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the pink plate", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/8e172cea37de7624b50db0e4.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0119/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 119, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 3, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 234, "role": "error_localization"}}
{"state": {"task": "to clean the table I need you to stack the JELLO box on pop corn box, and chips cylinder on JELLO box, stack pink plate on the other plate, put the croissant on top of pink plate", "plan": [{"step": 1, "subtask": "grasp JELLO box"}, {"step": 2, "subtask": "move grasped object to pop corn box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp chips cylinder"}, {"step": 5, "subtask": "move grasped object to JELLO box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink plate"}, {"step": 8, "subtask": "move grasped object to navy plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp croissant"}, {"step": 11, "subtask": "move grasped object to pink plate"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/8e172cea37de7624b50db0e4.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0120/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 120, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 3, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 419, "role": "error_localization"}}
{"state": {"task": "stack the blue plate on the pink plate, then put the banana and croissant on the blue plate, and finally lift up the blue plate for serving", "plan": [{"step": 1, "subtask": "grasp blue plate"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to blue plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp croissant"}, {"step": 8, "subtask": "move grasped object to blue plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp blue plate"}, {"step": 11, "subtask": "move grasped object up"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/8e172cea37de7624b50db0e4.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0121/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 121, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 3, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 386, "role": "error_localization"}}
{"state": {"task": "put the banana in the pink plate, then the pop corn box in the blue plate, then flip upright the chips cylinder, then swap the banana and pop corn box of plate", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pop corn box"}, {"step": 5, "subtask": "move grasped object to blue plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp chips cylinder"}, {"step": 8, "subtask": "move grasped object upright"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped object to blue plate"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pop corn box"}, {"step": 14, "subtask": "move grasped object to pink plate"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/d16305d9827d7935065b5e65.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0122/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/8", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 122, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 8, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 457, "role": "error_localization"}}
{"state": {"task": "put the banana in the pink plate, then the pop corn box in the blue plate, then flip upright the chips cylinder, then swap the banana and pop corn box of plate", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pop corn box"}, {"step": 5, "subtask": "move grasped object to blue plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped object to blue plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pop corn box"}, {"step": 11, "subtask": "move grasped object to pink plate"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/d16305d9827d7935065b5e65.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0123/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/8", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 123, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 8, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 408, "role": "error_localization"}}
{"state": {"task": "put the banana in the pink plate, then the pop corn box in the blue plate, then flip upright the chips cylinder, then swap the banana and pop corn box of plate", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp chips cylinder"}, {"step": 5, "subtask": "move grasped object upright"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pop corn box"}, {"step": 8, "subtask": "move grasped object to blue plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped object to blue plate"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pop corn box"}, {"step": 14, "subtask": "move grasped object to pink plate"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/d16305d9827d7935065b5e65.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0124/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/8", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 124, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 8, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 457, "role": "error_localization"}}
{"state": {"task": "stack the pink plate on blue plate, then put every other object on top of the pink plate", "plan": [{"step": 1, "subtask": "grasp pink plate"}, {"step": 2, "subtask": "move grasped object to blue plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to pink plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pop corn box"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp chips cylinder"}, {"step": 11, "subtask": "move grasped object to pink plate"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp JELLO box"}, {"step": 14, "subtask": "move grasped object to pink plate"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp blue cup"}, {"step": 17, "subtask": "move grasped object to pink plate"}, {"step": 18, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/935a5d8bf30047930c7b31f9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0125/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 125, "taskvar": "real_healthiest_in_plate+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 495, "role": "error_localization"}}
{"state": {"task": "put into the pink plate the banana, pop corn box, and JELLO box, then put into the navy plate the blue cup and chips cylinder", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pop corn box"}, {"step": 5, "subtask": "move grasped object to pink plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp JELLO box"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp blue cup"}, {"step": 11, "subtask": "move grasped object to navy plate"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp chips cylinder"}, {"step": 14, "subtask": "move grasped object to navy plate"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/935a5d8bf30047930c7b31f9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0126/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 126, "taskvar": "real_healthiest_in_plate+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "stack the pink plate on blue plate, then put every other object on top of the pink plate", "plan": [{"step": 1, "subtask": "grasp pink plate"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to pink plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pop corn box"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp chips cylinder"}, {"step": 11, "subtask": "move grasped object to pink plate"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp JELLO box"}, {"step": 14, "subtask": "move grasped object to pink plate"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp blue cup"}, {"step": 17, "subtask": "move grasped object to pink plate"}, {"step": 18, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/935a5d8bf30047930c7b31f9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0127/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 127, "taskvar": "real_healthiest_in_plate+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 495, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the blue plate then into the pink plate then next to the chips box.", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "release"}, {"step": 3, "subtask": "move grasped object to blue plate"}, {"step": 4, "subtask": "release"}, {"step": 5, "subtask": "grasp banana"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "move grasped banana to the pink plate"}, {"step": 8, "subtask": "release"}, {"step": 9, "subtask": "grasp banana"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "move grasped banana next to the chips box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/935a5d8bf30047930c7b31f9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0128/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 128, "taskvar": "real_healthiest_in_plate+0", "episode_id": 2, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 384, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the pink plate", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/a57594f270ccd1eaac281bf9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0129/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/3", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 129, "taskvar": "real_healthiest_in_plate+0", "episode_id": 3, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 234, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the pink plate, then remove it and put it on the table, oh, and finally stack the blue plate on the pink plate, then flip upright the chips cylinder and pop corn box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp blue plate"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp chips cylinder"}, {"step": 11, "subtask": "move grasped object upright"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pop corn box"}, {"step": 14, "subtask": "move grasped object to pink plate"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e311d28c8a7bc756786d5c24.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0130/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/4", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 130, "taskvar": "real_healthiest_in_plate+0", "episode_id": 4, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 463, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the pink plate, then remove it and put it on the table, oh, and finally stack the blue plate on the pink plate, then flip upright the chips cylinder and pop corn box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp chips cylinder"}, {"step": 8, "subtask": "move grasped object upright"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pop corn box"}, {"step": 11, "subtask": "move grasped object to pink plate"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e311d28c8a7bc756786d5c24.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0131/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/4", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 131, "taskvar": "real_healthiest_in_plate+0", "episode_id": 4, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 412, "role": "error_localization"}}
{"state": {"task": "put the pop corn box into the pink plate, grasp the healthiest food, lift it up, then put it in the blue plate, finally put the chips cylinder next to the pop corn box inside the pink plate", "plan": [{"step": 1, "subtask": "grasp pop corn box"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped banana up"}, {"step": 6, "subtask": "move grasped banana to blue plate"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp chips cylinder"}, {"step": 9, "subtask": "move grasped object next to the pop corn box inside the pink plate"}, {"step": 10, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/e311d28c8a7bc756786d5c24.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0132/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/4", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 132, "taskvar": "real_healthiest_in_plate+0", "episode_id": 4, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 390, "role": "error_localization"}}
{"state": {"task": "put in the pink plate the healthiest food you see on the table, then put it back on the table, then put it in the blue plate, and finally back to the pink plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped object to table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped object to blue plate"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/6abd8f8fa1b9d6d5298c7f4b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0133/ur5fail_plan_correct", "group_id": "ur5fail/real_healthiest_in_plate+0/5", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 133, "taskvar": "real_healthiest_in_plate+0", "episode_id": 5, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 406, "role": "error_localization"}}
{"state": {"task": "put the fruit in the plate on the napkin", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0134/ur5fail_plan_correct", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 134, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 235, "role": "error_localization"}}
{"state": {"task": "put the fruit in the plate that has a yellow cup next to it, then to the plate that has a navy cup next to it, finally to the plate that has a spoon next to it", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to navy plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to cyan plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0135/ur5fail_plan_correct", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 135, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "put the fruit in the plate that has a yellow cup next to it, then to the plate that has a navy cup next to it, finally to the plate that has a spoon next to it", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to yellow plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to navy plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0136/ur5fail_plan_correct", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 136, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 358, "role": "error_localization"}}
{"state": {"task": "put the fruit in the plate that has a spoon next to it, then grasp the plate that has a navy cup next to it and move the plate to the first plate manipulated", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp navy plate"}, {"step": 5, "subtask": "move grasped object to pink plate"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0137/ur5fail_plan_correct", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 137, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 308, "role": "error_localization"}}
{"state": {"task": "stack the yellow cup on top of the navy cup, then stack the pink plate on top of the cyan plate, then put the navy plate on to of the plate stack", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of navy cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pink plate"}, {"step": 5, "subtask": "move grasped object to top of cyan plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp navy plate"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": true, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0138/ur5fail_plan_correct", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 138, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "success", "failure_mode_raw": "ground_truth", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 360, "role": "error_localization"}}
{"state": {"task": "put the spoon sequentially in every plate of the table, then back to the table", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon into pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon into yellow plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon into cyan plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon to table"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "noul", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. Is this plan correct for the objective?", "label": false, "criteria": {"true": "Correct: executing the subtasks in order accomplishes the objective in this scene - every needed subtask is present, each one acts on the right object with the right target location or state, the order respects the dependencies, and no subtasks contradict each other.", "false": "Incorrect: the plan has at least one problem - a missing subtask, a wrong object, a wrong target location or state, a wrong order, or contradictory subtasks."}, "src": "ur5fail_plan_correct"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0139/ur5fail_plan_correct", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_correct", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 139, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 384, "role": "error_localization"}}
{"state": {"task": "move the ball to the right of the banana then move the hammer to the left of the banana, then swap the placement of the hammer and the ball while keeping the ball in the middle", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped ball to the right of the banana"}, {"step": 3, "subtask": "release ball"}, {"step": 4, "subtask": "grasp hammer"}, {"step": 5, "subtask": "move grasped hammer to the left of the banana"}, {"step": 6, "subtask": "release hammer"}, {"step": 7, "subtask": "move grasped ball to the left of the hammer"}, {"step": 8, "subtask": "grasp the ball"}, {"step": 9, "subtask": "release the ball"}, {"step": 10, "subtask": "grasp the hammer"}, {"step": 11, "subtask": "move grasped hammer to the right of the banana"}, {"step": 12, "subtask": "release hammer"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/ad4501b9673211a57736fdcf.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0001/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_pick_tool+0/173817592039", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 1, "taskvar": "real_pick_tool+0", "episode_id": 173817592039, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 454, "role": "error_localization"}}
{"state": {"task": "place the pizza to the box, then remove it, then put the strawberry in the box, then put the donut next to the corn", "plan": [{"step": 1, "subtask": "grasp pizza"}, {"step": 2, "subtask": "move grasped pizza to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pizza"}, {"step": 5, "subtask": "move grasped pizza out"}, {"step": 6, "subtask": "move grasped pizza to the table"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp donut"}, {"step": 9, "subtask": "move grasped donut next to the corn"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp strawberry"}, {"step": 12, "subtask": "move grasped strawberry to the box"}, {"step": 13, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0003/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 3, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 447, "role": "error_localization"}}
{"state": {"task": "place the pizza to the box, then remove it, then put the strawberry in the box, then put the donut next to the corn", "plan": [{"step": 1, "subtask": "grasp pizza"}, {"step": 2, "subtask": "move grasped pizza to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pizza"}, {"step": 5, "subtask": "move grasped pizza out"}, {"step": 6, "subtask": "move grasped pizza to the table"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp strawberry"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "move grasped strawberry to the box"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp donut"}, {"step": 13, "subtask": "move grasped donut next to the corn"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0004/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 4, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 461, "role": "error_localization"}}
{"state": {"task": "place the pizza to the box, then remove it, then put the strawberry in the box, then put the donut next to the corn", "plan": [{"step": 1, "subtask": "grasp pizza"}, {"step": 2, "subtask": "move grasped pizza to the box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "move grasped pizza out"}, {"step": 5, "subtask": "move grasped pizza to the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to the box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp donut"}, {"step": 11, "subtask": "move grasped donut next to the corn"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0005/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 5, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 431, "role": "error_localization"}}
{"state": {"task": "put the strawberry to the upper right part of the table, then to the bottom left part of the table, then to the upper left part of the table, then to the bottom right part", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the upper right part of the table"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped strawberry to the upper left part of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to the bottom left part of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped strawberry to the bottom right part"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0006/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 6, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 453, "role": "error_localization"}}
{"state": {"task": "put the strawberry to the upper right part of the table, then to the bottom left part of the table, then to the upper left part of the table, then to the bottom right part", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the upper right part of the table"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped strawberry to the bottom left part of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to the bottom right part"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0007/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 7, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 398, "role": "error_localization"}}
{"state": {"task": "put the strawberry to the upper right part of the table, then to the bottom left part of the table, then to the upper left part of the table, then to the bottom right part", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to the upper right part of the table"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped strawberry to the bottom left part of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped strawberry to the upper right part of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped strawberry to the bottom right part"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0008/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 8, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 453, "role": "error_localization"}}
{"state": {"task": "put 3 objects into the box", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp donut"}, {"step": 5, "subtask": "move grasped donut to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp corn"}, {"step": 8, "subtask": "release"}, {"step": 9, "subtask": "move grasped corn to box"}, {"step": 10, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0009/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 9, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 368, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box then the donut in the box then the pizza to the right of the corn", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp chicken"}, {"step": 5, "subtask": "move grasped chicken to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pizza"}, {"step": 8, "subtask": "move grasped pizza to the right of the corn"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0010/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 10, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 371, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the box then the donut in the box then the pizza to the right of the corn", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped strawberry to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pizza"}, {"step": 5, "subtask": "move grasped pizza to the right of the corn"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp donut"}, {"step": 8, "subtask": "move grasped donut to box"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0011/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 11, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 373, "role": "error_localization"}}
{"state": {"task": "put the tuna can in the box then put the corn in the box", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "move grasped object to box"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/39e1c6b4977d60d547029c82.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0012/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 12, "taskvar": "real_put_food_in_box+0", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 299, "role": "error_localization"}}
{"state": {"task": "put the banana then the black grapes in the box", "plan": [{"step": 1, "subtask": "move grasped banana to box"}, {"step": 2, "subtask": "release"}, {"step": 3, "subtask": "grasp black grapes"}, {"step": 4, "subtask": "move grasped object to box"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0016/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 16, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 295, "role": "error_localization"}}
{"state": {"task": "put the banana then the black grapes in the box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black grapes"}, {"step": 5, "subtask": "release"}, {"step": 6, "subtask": "move grasped object to box"}, {"step": 7, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0017/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 17, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 323, "role": "error_localization"}}
{"state": {"task": "put all the fruits you see next to the strawberry", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped banana to strawberry"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black grapes"}, {"step": 5, "subtask": "move grasped black grapes to strawberry"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "move grasped green pear to strawberry"}, {"step": 8, "subtask": "release"}, {"step": 9, "subtask": "grasp purple plum"}, {"step": 10, "subtask": "move grasped purple plum to strawberry"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp yellow peach"}, {"step": 13, "subtask": "move grasped yellow peach to strawberry"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/54fb54762906d1dd99b74a7d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0023/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 23, "taskvar": "real_put_fruit_in_box+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 444, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "move grasped object out"}, {"step": 8, "subtask": "move grasped lemon to table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped banana to box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0029/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 29, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 417, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp peach"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped lemon to table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp banana"}, {"step": 12, "subtask": "move grasped banana to box"}, {"step": 13, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0030/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 30, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 433, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box, remove the lemon, then put the banana in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped lemon to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp peach"}, {"step": 5, "subtask": "move grasped peach to box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped lemon to table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp banana"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "move grasped banana to box"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0031/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 31, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 447, "role": "error_localization"}}
{"state": {"task": "put the lemon then the peach in the box", "plan": [{"step": 1, "subtask": "grasp lemon"}, {"step": 2, "subtask": "move grasped object to box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "move grasped object to box"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/19791c4069a9720a86491d0d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0032/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_fruit_in_box+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 32, "taskvar": "real_put_fruit_in_box+3", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 293, "role": "error_localization"}}
{"state": {"task": "put the ball in the top drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped object to top drawer"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp handle of the top drawer"}, {"step": 5, "subtask": "move grasped object to closed drawer position"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0040/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 40, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 318, "role": "error_localization"}}
{"state": {"task": "put the ball in the top drawer", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped object to top drawer"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0041/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 41, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 261, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp ball"}, {"step": 2, "subtask": "move grasped object to bottom drawer"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pink cube"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp frog toy"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp handle of the bottom drawer"}, {"step": 11, "subtask": "move grasped object to closed drawer position"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0043/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 43, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 423, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp handle of the bottom drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cube"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp frog toy"}, {"step": 11, "subtask": "move grasped object to top drawer"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp handle of the bottom drawer"}, {"step": 14, "subtask": "move grasped object to closed drawer position"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0044/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 44, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 478, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp handle of the bottom drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cube"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp frog toy"}, {"step": 11, "subtask": "move grasped object to bottom drawer"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp handle of the middle drawer"}, {"step": 14, "subtask": "move grasped object to closed drawer position"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0045/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 45, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 478, "role": "error_localization"}}
{"state": {"task": "put the ball, pink cube, frog toy in the bottom drawer, then close the drawer", "plan": [{"step": 1, "subtask": "grasp handle of the bottom drawer"}, {"step": 2, "subtask": "move grasped object to open drawer position"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to bottom drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cube"}, {"step": 8, "subtask": "move grasped object to bottom drawer"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "move grasped object to bottom drawer"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp handle of the bottom drawer"}, {"step": 13, "subtask": "move grasped object to closed drawer position"}, {"step": 14, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0046/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 46, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 461, "role": "error_localization"}}
{"state": {"task": "put ball in top drawer, close it, then put pink cube in middle drawer and close it", "plan": [{"step": 1, "subtask": "grasp handle of the top drawer"}, {"step": 2, "subtask": "move grasped object out"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to top drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp handle of the top drawer"}, {"step": 8, "subtask": "move grasped object to closed drawer position"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp handle of the middle drawer"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "move grasped object to open drawer position"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp pink cube"}, {"step": 15, "subtask": "move grasped object to middle drawer"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp handle of the middle drawer"}, {"step": 18, "subtask": "move grasped object to closed drawer position"}, {"step": 19, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0049/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 49, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 549, "role": "error_localization"}}
{"state": {"task": "put ball in top drawer, close it, then put pink cube in middle drawer and close it", "plan": [{"step": 1, "subtask": "grasp handle of the top drawer"}, {"step": 2, "subtask": "move grasped object out"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to top drawer"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp handle of the top drawer"}, {"step": 8, "subtask": "move grasped object to closed drawer position"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp handle of the middle drawer"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "move grasped object to open drawer position"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp pink cube"}, {"step": 15, "subtask": "move grasped object to middle drawer"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp handle of the top drawer"}, {"step": 18, "subtask": "move grasped object to closed drawer position"}, {"step": 19, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0050/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 50, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 549, "role": "error_localization"}}
{"state": {"task": "put the ball in the top part of the drawer", "plan": [{"step": 1, "subtask": "grasp top part of the drawer"}, {"step": 2, "subtask": "release"}, {"step": 3, "subtask": "grasp ball"}, {"step": 4, "subtask": "move grasped object to top part of the drawer"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/44a5a6c9f2f1d06ea6032859.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0051/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_in_drawer+2/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 51, "taskvar": "real_put_item_in_drawer+2", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 299, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the plate, then the tuna can on the upper right of the table, then the tuna can on the lower right of the table, then remove the strawberry from the plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tuna can"}, {"step": 5, "subtask": "move grasped object to upper right of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped object to lower right of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped object to table"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/23196e5cafe425f2d66448cd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0053/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 53, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 441, "role": "error_localization"}}
{"state": {"task": "put the strawberry in the plate, then the tuna can on the upper right of the table, then the tuna can on the lower right of the table, then remove the strawberry from the plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp tuna can"}, {"step": 5, "subtask": "move grasped object to upper right of the table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp tuna can"}, {"step": 8, "subtask": "move grasped object to lower right of the table"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp tuna can"}, {"step": 11, "subtask": "move grasped object to table"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/23196e5cafe425f2d66448cd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0054/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 54, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 443, "role": "error_localization"}}
{"state": {"task": "pick the green pear and place it next to the ball", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object to container"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/66f558ab24ef777dc5829b87.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0062/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816518206", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 62, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816518206, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 265, "role": "error_localization"}}
{"state": {"task": "pick the green pear and place it next to box", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object into the wooden box box"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/799fe8e2dfc95526aa62ae73.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0063/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816539090", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 63, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816539090, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 267, "role": "error_localization"}}
{"state": {"task": "put into the box all the objects that are on the table, but before that, if you see any object inside the box, remove it", "plan": [{"step": 1, "subtask": "grasp green pear"}, {"step": 2, "subtask": "move grasped object out"}, {"step": 3, "subtask": "move grasped object to the table"}, {"step": 4, "subtask": "release"}, {"step": 5, "subtask": "grasp hot dog"}, {"step": 6, "subtask": "move grasped object to wooden box"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp chips container"}, {"step": 9, "subtask": "move grasped object to wooden box"}, {"step": 10, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/346b0c0621b7ca68f0b77d98.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0066/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816553691", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 66, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816553691, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 397, "role": "error_localization"}}
{"state": {"task": "place the healthiest food you see in the wooden box", "plan": [{"step": 1, "subtask": "grasp chocolate"}, {"step": 2, "subtask": "move grasped object to the wooden box"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/98405d9e80f43cbc72e8a360.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0067/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173816578860", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 67, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173816578860, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 266, "role": "error_localization"}}
{"state": {"task": "flip upright the bottle, then put the bread to the left of the bottle, then put the bread to the right of the bottle, then on top of the pop corn box", "plan": [{"step": 1, "subtask": "grasp bottle"}, {"step": 2, "subtask": "rotate grasped object"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp bread"}, {"step": 5, "subtask": "move grasped object to the right of the bottle"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp bread"}, {"step": 8, "subtask": "move grasped object to the left of the bottle"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp bread"}, {"step": 11, "subtask": "move grasped object to the top of pop corn box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/a721c33537bfdecba849aa6d.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0069/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817342016", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 69, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817342016, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 440, "role": "error_localization"}}
{"state": {"task": "pick up the healthiest food, then move it to each object of the table sequentially, finally put it next to the coke bottle", "plan": [{"step": 1, "subtask": "grasp carrot"}, {"step": 2, "subtask": "move grasped object to the coke bottle"}, {"step": 3, "subtask": "move grasped object to the bread"}, {"step": 4, "subtask": "move grasped object next to the coke bottle"}, {"step": 5, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/138f5eaf01013f703f20adfe.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0070/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_item_spatial_in_plate+0/173817351532", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 70, "taskvar": "real_put_item_spatial_in_plate+0", "episode_id": 173817351532, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 321, "role": "error_localization"}}
{"state": {"task": "stack the blue cup on top of the red cube, then put the ball inside the blue cup, then unstack the tower without making anything fall", "plan": [{"step": 1, "subtask": "grasp blue cup"}, {"step": 2, "subtask": "move grasped object to top of the red cube"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object into the blue cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp ball"}, {"step": 8, "subtask": "move grasped object out"}, {"step": 9, "subtask": "move grasped object to the table"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "grasp red cube"}, {"step": 12, "subtask": "move grasped object to the table"}, {"step": 13, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/458ccd670843141efd784cbd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0071/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_sport+0/173817680775", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 71, "taskvar": "real_sport+0", "episode_id": 173817680775, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 451, "role": "error_localization"}}
{"state": {"task": "move next to the carrot every object you see", "plan": [{"step": 1, "subtask": "grasp blue cup"}, {"step": 2, "subtask": "move grasped object to next to the carrot"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to next to the carrot"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp carrot"}, {"step": 8, "subtask": "move grasped object to next to the carrot"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/458ccd670843141efd784cbd.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0073/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_sport+0/173817680775", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 73, "taskvar": "real_sport+0", "episode_id": 173817680775, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 364, "role": "error_localization"}}
{"state": {"task": "stack a pink cup onto a black cup onto a cyan cup onto a yellow cup onto a navy cup", "plan": [{"step": 1, "subtask": "grasp pink cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black cup"}, {"step": 5, "subtask": "move grasped object to top of cyan cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp cyan cup"}, {"step": 8, "subtask": "move grasped object to top of navy cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp navy cup"}, {"step": 11, "subtask": "move grasped object to top of yellow cup"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0074/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 74, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 430, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of navy cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp cyan cup"}, {"step": 5, "subtask": "move grasped object to top of yellow cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp black cup"}, {"step": 8, "subtask": "move grasped object to top of cyan cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0075/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 75, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 370, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of navy cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black cup"}, {"step": 5, "subtask": "move grasped object to top of cyan cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cup"}, {"step": 8, "subtask": "move grasped object to top of black cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0076/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 76, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 370, "role": "error_localization"}}
{"state": {"task": "create a stack of four cups: yellow on navy on pink on black cups", "plan": [{"step": 1, "subtask": "grasp pink cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp navy cup"}, {"step": 5, "subtask": "move grasped object to top of yellow cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp yellow cup"}, {"step": 8, "subtask": "move grasped object to top of pink cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0077/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 77, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 372, "role": "error_localization"}}
{"state": {"task": "create two stacks of cups: one with the yellow and navy, then one with the cyan, black and pink cups", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of navy cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp black cup"}, {"step": 5, "subtask": "move grasped object to top of pink cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp cyan cup"}, {"step": 8, "subtask": "move grasped object to top of navy cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0078/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 78, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 380, "role": "error_localization"}}
{"state": {"task": "create two stacks of cups: one with the yellow and navy, then one with the cyan, and pink cups", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "release"}, {"step": 3, "subtask": "move grasped object to top of navy cup"}, {"step": 4, "subtask": "release"}, {"step": 5, "subtask": "grasp cyan cup"}, {"step": 6, "subtask": "move grasped object to top of pink cup"}, {"step": 7, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0079/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 79, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 342, "role": "error_localization"}}
{"state": {"task": "create a stack of cups: navy on top of pink, pink on top of cyan, cyan on top of yellow, yellow on top of black", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp cyan cup"}, {"step": 5, "subtask": "move grasped object to top of pink cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pink cup"}, {"step": 8, "subtask": "move grasped object to top of cyan cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp navy cup"}, {"step": 11, "subtask": "move grasped object to top of pink cup"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/77f8b26c7e3e436a17c54551.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0081/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+3/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 81, "taskvar": "real_stack_cup+3", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 439, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table", "plan": [{"step": 1, "subtask": "grasp black cup"}, {"step": 2, "subtask": "move grasped object to top of red cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of black cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of yellow cup"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0083/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 83, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 370, "role": "error_localization"}}
{"state": {"task": "stack together in one tower all the cups you see on the table, except the cyan cup that should be moved next but not on top of the stack", "plan": [{"step": 1, "subtask": "grasp black cup"}, {"step": 2, "subtask": "move grasped object to top of red cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of green cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of yellow cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp cyan cup"}, {"step": 11, "subtask": "move grasped object next to the stack"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0085/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 85, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 439, "role": "error_localization"}}
{"state": {"task": "stack a black cup onto a red cup", "plan": [{"step": 1, "subtask": "grasp red cup"}, {"step": 2, "subtask": "move grasped object to black cup"}, {"step": 3, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0086/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 86, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 263, "role": "error_localization"}}
{"state": {"task": "put the red cup on top of black cup, then put the yellow cup on top of cyan cup, then put the green cup on top of the red cup, then unstack the stack containing the red cup", "plan": [{"step": 1, "subtask": "grasp red cup"}, {"step": 2, "subtask": "move grasped object to top of black cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp yellow cup"}, {"step": 5, "subtask": "move grasped object to top of cyan cup"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green cup"}, {"step": 8, "subtask": "move grasped object to top of cyan cup"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp green cup"}, {"step": 11, "subtask": "move grasped object out"}, {"step": 12, "subtask": "move grasped object to table"}, {"step": 13, "subtask": "release"}, {"step": 14, "subtask": "grasp red cup"}, {"step": 15, "subtask": "move grasped object out"}, {"step": 16, "subtask": "move grasped object to table"}, {"step": 17, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/84be412c8fe4869ed0a2680f.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0087/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup+4/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 87, "taskvar": "real_stack_cup+4", "episode_id": 0, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 535, "role": "error_localization"}}
{"state": {"task": "grasp the cup that is the closest to the red apple, and put it on the pink cup, then unstack them", "plan": [{"step": 1, "subtask": "grasp navy cup"}, {"step": 2, "subtask": "move grasped object to the pink cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp navy cup"}, {"step": 5, "subtask": "move grasped object out"}, {"step": 6, "subtask": "move grasped object to table"}, {"step": 7, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/6af2e641d831d29db4c016c2.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0092/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup_spatial+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 92, "taskvar": "real_stack_cup_spatial+0", "episode_id": 0, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 345, "role": "error_localization"}}
{"state": {"task": "grasp the cup that is the closest to the red apple, and put it on the pink cup, then put the apple on top of the stack, then unstack the entire stack", "plan": [{"step": 1, "subtask": "grasp yellow cup"}, {"step": 2, "subtask": "move grasped object to the pink cup"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "release"}, {"step": 6, "subtask": "move grasped object to top of yellow cup"}, {"step": 7, "subtask": "release"}, {"step": 8, "subtask": "grasp red apple"}, {"step": 9, "subtask": "move grasped object out"}, {"step": 10, "subtask": "move grasped object to table"}, {"step": 11, "subtask": "release"}, {"step": 12, "subtask": "grasp yellow cup"}, {"step": 13, "subtask": "move grasped object out"}, {"step": 14, "subtask": "move grasped object to table"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/6af2e641d831d29db4c016c2.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0093/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_stack_cup_spatial+0/0", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 93, "taskvar": "real_stack_cup_spatial+0", "episode_id": 0, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 490, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the strawberry then the banana, then the lemon then the yellow peach, then put them back on the table", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp yellow peach"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp lemon"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp strawberry"}, {"step": 14, "subtask": "move grasped object out"}, {"step": 15, "subtask": "move grasped object to table"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp banana"}, {"step": 18, "subtask": "move grasped object out"}, {"step": 19, "subtask": "move grasped object to table"}, {"step": 20, "subtask": "release"}, {"step": 21, "subtask": "grasp lemon"}, {"step": 22, "subtask": "move grasped object out"}, {"step": 23, "subtask": "move grasped object to table"}, {"step": 24, "subtask": "release"}, {"step": 25, "subtask": "grasp yellow peach"}, {"step": 26, "subtask": "move grasped object out"}, {"step": 27, "subtask": "move grasped object to table"}, {"step": 28, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e84eaff8c1b974397c2a3d23.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0098/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 98, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 694, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the strawberry then the banana, then the lemon then the yellow peach, then put them back on the table", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow corn"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp strawberry"}, {"step": 14, "subtask": "move grasped object out"}, {"step": 15, "subtask": "move grasped object to table"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp banana"}, {"step": 18, "subtask": "move grasped object out"}, {"step": 19, "subtask": "move grasped object to table"}, {"step": 20, "subtask": "release"}, {"step": 21, "subtask": "grasp lemon"}, {"step": 22, "subtask": "move grasped object out"}, {"step": 23, "subtask": "move grasped object to table"}, {"step": 24, "subtask": "release"}, {"step": 25, "subtask": "grasp yellow peach"}, {"step": 26, "subtask": "move grasped object out"}, {"step": 27, "subtask": "move grasped object to table"}, {"step": 28, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e84eaff8c1b974397c2a3d23.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0099/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 99, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 694, "role": "error_localization"}}
{"state": {"task": "put into the wooden box the strawberry then the banana, then the lemon then the yellow peach, then put them back on the table", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp lemon"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp strawberry"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "move grasped object out"}, {"step": 16, "subtask": "move grasped object to table"}, {"step": 17, "subtask": "release"}, {"step": 18, "subtask": "grasp banana"}, {"step": 19, "subtask": "move grasped object out"}, {"step": 20, "subtask": "move grasped object to table"}, {"step": 21, "subtask": "release"}, {"step": 22, "subtask": "grasp lemon"}, {"step": 23, "subtask": "move grasped object out"}, {"step": 24, "subtask": "move grasped object to table"}, {"step": 25, "subtask": "release"}, {"step": 26, "subtask": "grasp yellow peach"}, {"step": 27, "subtask": "move grasped object out"}, {"step": 28, "subtask": "move grasped object to table"}, {"step": 29, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e84eaff8c1b974397c2a3d23.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0100/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/3", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 100, "taskvar": "real_put_food_in_box+0", "episode_id": 3, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 708, "role": "error_localization"}}
{"state": {"task": "put into the wooden box all the objects you see on the table", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp yellow corn cob"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "move grasped object to wooden box"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp strawberry"}, {"step": 18, "subtask": "move grasped object to wooden box"}, {"step": 19, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/bd0f90f8f9fb55c990e0d017.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0105/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 105, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 531, "role": "error_localization"}}
{"state": {"task": "put into the wooden box all the objects you see on the table, then put them all back on the table", "plan": [{"step": 1, "subtask": "grasp tuna can"}, {"step": 2, "subtask": "move grasped object to wooden box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp red apple"}, {"step": 5, "subtask": "move grasped object to wooden box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp green pear"}, {"step": 8, "subtask": "move grasped object to wooden box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp yellow peach"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp yellow corn cob"}, {"step": 14, "subtask": "move grasped object to wooden box"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp strawberry"}, {"step": 17, "subtask": "move grasped object to wooden box"}, {"step": 18, "subtask": "release"}, {"step": 19, "subtask": "grasp tuna can"}, {"step": 20, "subtask": "move grasped object out"}, {"step": 21, "subtask": "move grasped object to table"}, {"step": 22, "subtask": "release"}, {"step": 23, "subtask": "grasp red apple"}, {"step": 24, "subtask": "move grasped object out"}, {"step": 25, "subtask": "move grasped object to table"}, {"step": 26, "subtask": "release"}, {"step": 27, "subtask": "grasp green pear"}, {"step": 28, "subtask": "move grasped object out"}, {"step": 29, "subtask": "move grasped object to table"}, {"step": 30, "subtask": "release"}, {"step": 31, "subtask": "grasp yellow peach"}, {"step": 32, "subtask": "move grasped object to wooden box"}, {"step": 33, "subtask": "release"}, {"step": 34, "subtask": "grasp yellow corn cob"}, {"step": 35, "subtask": "move grasped object out"}, {"step": 36, "subtask": "move grasped object to table"}, {"step": 37, "subtask": "release"}, {"step": 38, "subtask": "grasp strawberry"}, {"step": 39, "subtask": "move grasped object out"}, {"step": 40, "subtask": "move grasped object to table"}, {"step": 41, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/bd0f90f8f9fb55c990e0d017.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0106/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_food_in_box+0/11", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 106, "taskvar": "real_put_food_in_box+0", "episode_id": 11, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 917, "role": "error_localization"}}
{"state": {"task": "clean the table by storing everything in the cardboard box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp green plate"}, {"step": 5, "subtask": "move grasped object to cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp white baseball ball"}, {"step": 8, "subtask": "move grasped object to cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to cardboard box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/47ff4890929bdb729765135b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0110/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/1", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 110, "taskvar": "real_put_carrot_in_container+0", "episode_id": 1, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 411, "role": "error_localization"}}
{"state": {"task": "clean the table by storing everything in the cardboard box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp green plate"}, {"step": 5, "subtask": "move grasped object to cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp white baseball ball"}, {"step": 8, "subtask": "move grasped object to cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp scissors"}, {"step": 11, "subtask": "move grasped object to cardboard box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pen"}, {"step": 14, "subtask": "release"}, {"step": 15, "subtask": "move grasped object to cardboard box"}, {"step": 16, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/47ff4890929bdb729765135b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0111/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/1", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 111, "taskvar": "real_put_carrot_in_container+0", "episode_id": 1, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 475, "role": "error_localization"}}
{"state": {"task": "put into the big cardboard box the banana, then the white baseball ball, then the scissors, then the pen, and finally put them all into the wooden box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to big cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp white baseball ball"}, {"step": 5, "subtask": "move grasped object to big cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to big cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped object to wooden box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp white baseball ball"}, {"step": 14, "subtask": "move grasped object to wooden box"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp scissors"}, {"step": 17, "subtask": "move grasped object to wooden box"}, {"step": 18, "subtask": "release"}, {"step": 19, "subtask": "grasp pen"}, {"step": 20, "subtask": "move grasped object to wooden box"}, {"step": 21, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0114/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 114, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 587, "role": "error_localization"}}
{"state": {"task": "put into the big cardboard box the banana, then the white baseball ball, then the scissors, then the pen, and finally put them all into the wooden box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to big cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp white baseball ball"}, {"step": 5, "subtask": "move grasped object to big cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to big cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to big cardboard box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp banana"}, {"step": 14, "subtask": "move grasped object to wooden box"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp white baseball ball"}, {"step": 17, "subtask": "move grasped object to wooden box"}, {"step": 18, "subtask": "release"}, {"step": 19, "subtask": "move grasped object to wooden box"}, {"step": 20, "subtask": "grasp scissors"}, {"step": 21, "subtask": "release"}, {"step": 22, "subtask": "grasp pen"}, {"step": 23, "subtask": "move grasped object to wooden box"}, {"step": 24, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0115/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 115, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 638, "role": "error_localization"}}
{"state": {"task": "put into the big cardboard box the banana, then the white baseball ball, then the scissors, then the pen, and finally put the pen and scissors into the wooden box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to big cardboard box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp white baseball ball"}, {"step": 5, "subtask": "move grasped object to big cardboard box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to big cardboard box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to big cardboard box"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pen"}, {"step": 14, "subtask": "move grasped object out"}, {"step": 15, "subtask": "move grasped object to wooden box"}, {"step": 16, "subtask": "release"}, {"step": 17, "subtask": "grasp scissors"}, {"step": 18, "subtask": "move grasped object out"}, {"step": 19, "subtask": "move grasped object to wooden box"}, {"step": 20, "subtask": "release"}, {"step": 21, "subtask": "grasp banana"}, {"step": 22, "subtask": "move grasped object out"}, {"step": 23, "subtask": "move grasped object to wooden box"}, {"step": 24, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0116/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 116, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 642, "role": "error_localization"}}
{"state": {"task": "put every object in the largest box (except the other box)", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to wooden square box"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp ball"}, {"step": 5, "subtask": "move grasped object to wooden square box"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp scissors"}, {"step": 8, "subtask": "move grasped object to wooden square box"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pen"}, {"step": 11, "subtask": "move grasped object to wooden square box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/28efb1ac85471ccec2714b4e.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0117/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_carrot_in_container+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 117, "taskvar": "real_put_carrot_in_container+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 414, "role": "error_localization"}}
{"state": {"task": "put the banana in the pink plate, then the pop corn box in the blue plate, then flip upright the chips cylinder, then swap the banana and pop corn box of plate", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp pop corn box"}, {"step": 5, "subtask": "move grasped object to blue plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped object to blue plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pop corn box"}, {"step": 11, "subtask": "move grasped object to pink plate"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/d16305d9827d7935065b5e65.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0123/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/8", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 123, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 8, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 437, "role": "error_localization"}}
{"state": {"task": "put the banana in the pink plate, then the pop corn box in the blue plate, then flip upright the chips cylinder, then swap the banana and pop corn box of plate", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp chips cylinder"}, {"step": 5, "subtask": "move grasped object upright"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pop corn box"}, {"step": 8, "subtask": "move grasped object to blue plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp banana"}, {"step": 11, "subtask": "move grasped object to blue plate"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp pop corn box"}, {"step": 14, "subtask": "move grasped object to pink plate"}, {"step": 15, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/d16305d9827d7935065b5e65.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0124/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_healthiest_banana_in_plate+0/8", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 124, "taskvar": "real_healthiest_banana_in_plate+0", "episode_id": 8, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 486, "role": "error_localization"}}
{"state": {"task": "stack the pink plate on blue plate, then put every other object on top of the pink plate", "plan": [{"step": 1, "subtask": "grasp pink plate"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to pink plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp pop corn box"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp chips cylinder"}, {"step": 11, "subtask": "move grasped object to pink plate"}, {"step": 12, "subtask": "release"}, {"step": 13, "subtask": "grasp JELLO box"}, {"step": 14, "subtask": "move grasped object to pink plate"}, {"step": 15, "subtask": "release"}, {"step": 16, "subtask": "grasp blue cup"}, {"step": 17, "subtask": "move grasped object to pink plate"}, {"step": 18, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/935a5d8bf30047930c7b31f9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0127/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_healthiest_in_plate+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 127, "taskvar": "real_healthiest_in_plate+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 524, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the blue plate then into the pink plate then next to the chips box.", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "release"}, {"step": 3, "subtask": "move grasped object to blue plate"}, {"step": 4, "subtask": "release"}, {"step": 5, "subtask": "grasp banana"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "move grasped banana to the pink plate"}, {"step": 8, "subtask": "release"}, {"step": 9, "subtask": "grasp banana"}, {"step": 10, "subtask": "release"}, {"step": 11, "subtask": "move grasped banana next to the chips box"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "contradictory_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/935a5d8bf30047930c7b31f9.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0128/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_healthiest_in_plate+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 128, "taskvar": "real_healthiest_in_plate+0", "episode_id": 2, "failure_category": "contradictory_subtasks", "failure_mode_raw": "contradictory subtasks", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 413, "role": "error_localization"}}
{"state": {"task": "put the healthiest food in the pink plate, then remove it and put it on the table, oh, and finally stack the blue plate on the pink plate, then flip upright the chips cylinder and pop corn box", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp chips cylinder"}, {"step": 8, "subtask": "move grasped object upright"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp pop corn box"}, {"step": 11, "subtask": "move grasped object to pink plate"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "missing_subtasks", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/e311d28c8a7bc756786d5c24.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0131/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_healthiest_in_plate+0/4", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 131, "taskvar": "real_healthiest_in_plate+0", "episode_id": 4, "failure_category": "missing_subtasks", "failure_mode_raw": "missing subtask", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 441, "role": "error_localization"}}
{"state": {"task": "put in the pink plate the healthiest food you see on the table, then put it back on the table, then put it in the blue plate, and finally back to the pink plate", "plan": [{"step": 1, "subtask": "grasp strawberry"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp strawberry"}, {"step": 5, "subtask": "move grasped object to table"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp strawberry"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp strawberry"}, {"step": 11, "subtask": "move grasped object to blue plate"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_order", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/6abd8f8fa1b9d6d5298c7f4b.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0133/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_healthiest_in_plate+0/5", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 133, "taskvar": "real_healthiest_in_plate+0", "episode_id": 5, "failure_category": "wrong_order", "failure_mode_raw": "wrong order", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 435, "role": "error_localization"}}
{"state": {"task": "put the fruit in the plate that has a yellow cup next to it, then to the plate that has a navy cup next to it, finally to the plate that has a spoon next to it", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to yellow plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp banana"}, {"step": 5, "subtask": "move grasped object to navy plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp banana"}, {"step": 8, "subtask": "move grasped object to pink plate"}, {"step": 9, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0136/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 136, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 387, "role": "error_localization"}}
{"state": {"task": "put the fruit in the plate that has a spoon next to it, then grasp the plate that has a navy cup next to it and move the plate to the first plate manipulated", "plan": [{"step": 1, "subtask": "grasp banana"}, {"step": 2, "subtask": "move grasped object to pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp navy plate"}, {"step": 5, "subtask": "move grasped object to pink plate"}, {"step": 6, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_object", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0137/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 137, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "wrong_object", "failure_mode_raw": "wrong object manipulated", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 337, "role": "error_localization"}}
{"state": {"task": "put the spoon sequentially in every plate of the table, then back to the table", "plan": [{"step": 1, "subtask": "grasp spoon"}, {"step": 2, "subtask": "move grasped spoon into pink plate"}, {"step": 3, "subtask": "release"}, {"step": 4, "subtask": "grasp spoon"}, {"step": 5, "subtask": "move grasped spoon into yellow plate"}, {"step": 6, "subtask": "release"}, {"step": 7, "subtask": "grasp spoon"}, {"step": 8, "subtask": "move grasped spoon into cyan plate"}, {"step": 9, "subtask": "release"}, {"step": 10, "subtask": "grasp spoon"}, {"step": 11, "subtask": "move grasped spoon to table"}, {"step": 12, "subtask": "release"}], "images": ["initial scene, before the robot acts"]}, "questions": {"decision": {"type": "choice", "instructions": "You are checking the plan of a UR5 robot arm with a parallel gripper before it acts. The state gives the high-level objective and the proposed plan: an ordered list of subtasks the robot will execute. The attached image shows the initial scene. This plan is known to be INCORRECT. Which category best describes its error?", "criteria": {"missing_subtasks": "Missing subtasks: one or several needed subtasks are missing.", "wrong_object": "Wrong object manipulated: one or several subtasks manipulate the wrong object.", "wrong_state_or_placement": "Wrong object state or placement: one or several subtasks select the wrong target, location or state for the correct object.", "wrong_order": "Wrong order: one or several subtasks are not in the right order, breaking causal dependencies.", "contradictory_subtasks": "Contradictory subtasks: some subtasks conflict with each other."}, "label": "wrong_state_or_placement", "src": "ur5fail_plan_failure_mode"}}, "media": [{"type": "image", "uri": "media/ur5fail/50dfc620e255ec6feba2203c.png"}], "_meta": {"source": "ur5fail", "origin": "paulpacaud/Guardian-FailCoT-OOD-datasets@bd2b24224e4da833a8d42ca694d839622e54bb48", "track": "public", "modality": "image", "variant": "clean", "license": "Apache-2.0", "id": "ur5fail/plan0139/ur5fail_plan_failure_mode", "group_id": "ur5fail/real_put_banana_in_plate+0/2", "family": "ur5fail_plan_failure_mode", "label_source": "construction", "extra": {"sub_benchmark": "UR5-Fail", "native_line": 139, "taskvar": "real_put_banana_in_plate+0", "episode_id": 2, "failure_category": "wrong_state_or_placement", "failure_mode_raw": "wrong object state or placement", "image_sizes": [[256, 256]], "media_policy": "upstream images copied byte-for-byte (original resolution and encoding), in Guardian's prompt order; initial scene only"}, "n_media": 1, "approx_tokens": 413, "role": "error_localization"}}